{ "data": { "similar": [ { "grade": "BB", "json": "https://www.anchorterminal.com/tools/github-mcp-server.json", "name": "GitHub MCP Server", "score": 70.4, "shared": [ "code.repo" ], "slug": "github-mcp-server" }, { "grade": "B", "json": "https://www.anchorterminal.com/tools/entire.json", "name": "Entire", "score": 69.4, "shared": [ "code.repo" ], "slug": "entire" }, { "grade": "B", "json": "https://www.anchorterminal.com/tools/graphify.json", "name": "Graphify", "score": 68.9, "shared": [ "code.repo" ], "slug": "graphify" }, { "grade": "B", "json": "https://www.anchorterminal.com/tools/azure-devops-mcp.json", "name": "Azure DevOps MCP Server", "score": 66.4, "shared": [ "code.repo" ], "slug": "azure-devops-mcp" }, { "grade": "C", "json": "https://www.anchorterminal.com/tools/atlassian-rovo-mcp.json", "name": "Atlassian Rovo MCP Server", "score": 57.9, "shared": [ "code.repo" ], "slug": "atlassian-rovo-mcp" }, { "grade": "C", "json": "https://www.anchorterminal.com/tools/sourcegraph-mcp.json", "name": "Sourcegraph MCP Server", "score": 57.5, "shared": [ "code.repo" ], "slug": "sourcegraph-mcp" } ], "tool": { "slug": "tokeneconomizer", "name": "TokenEconomizer", "vendor": "Microsoft", "vendorUrl": "https://tokeneconomizer.github.io/", "kind": "framework", "category": "code", "summary": "TokenEconomizer is an Apache-2.0 plugin from Microsoft Research and UC Berkeley that cuts repeat coding-agent cost in a repository by injecting learned facts, pruning unused tools and redirecting stuck turns. It runs in Claude Code, Codex, Copilot CLI and Pi.", "url": "https://www.anchorterminal.com/tools/tokeneconomizer", "markdownUrl": "https://www.anchorterminal.com/tools/tokeneconomizer.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/tokeneconomizer.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/tokeneconomizer.json", "repo": "https://github.com/microsoft/TokenEconomizer", "license": "Apache-2.0", "transports": [], "packages": [], "auth": "none", "authNotes": "No account, key or login of its own. It runs inside Claude Code, Codex, the Copilot CLI or Pi, and its own model calls go through that agent's CLI on whatever plan or login the person already has.", "pricing": "free", "pricingNotes": "Free under Apache 2.0, with nothing to buy. Its loop-guard judge, the redirect writer and the optional micro-skills proposer make model calls billed on the person's agent plan, Haiku by default on Claude Code, and `te status` reports that cost.", "priceSummary": "Free · OSS", "where": "library", "x402": { "level": "no", "evidence": "Local software with nothing to buy. No x402, MPP or L402 in the repository or on the project page, checked 30 October 2026.", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": 976, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-30" }, "docsUrl": "https://github.com/microsoft/TokenEconomizer/blob/main/plugin/README.md", "capabilities": [ "code.repo" ], "tags": [ "open-source", "apache-2.0", "local", "free", "python", "claude-code-plugin", "hooks", "pre-1.0" ], "graded": true, "anchor": { "graded": true, "score": 62.6, "grade": "B", "agentReady": false, "rank": 502, "ranked": true, "rankOf": 1244, "categoryRank": 9, "methodology": "0.7", "run": "2026-10-01", "scores": { "ergonomics": 76, "maintenance": 43, "payments": 60, "reliability": 46, "schema": 66, "security": 71, "transparency": 76 }, "pending": [ "performance", "tasks" ], "breakdown": [ { "key": "reliability", "name": "Reliability", "weight": 16, "effectiveWeight": 20, "score": 46, "points": 9.2, "reason": "Local-package reading, as for superpowers and impeccable. Installs from a git clone through each agent's own plugin manager (a local marketplace on Claude Code, Codex and Copilot, `pi install git:` on Pi), not from a registry or an official marketplace, with Python 3.10 or later and a supported-versions table per agent (14 of 20). .github/workflows/tests.yml runs pytest on push to main and on pull requests, on Ubuntu and Windows with Python 3.10 and 3.13, plus a Codex end-to-end job pinned to 0.164.0. 17 test files in tests/. Whether the runs pass wasn't checked, as github.com couldn't be read (14 of 25). The archived repository page of 30 October 2026 shows 1 open issue and 5 open pull requests, few for a project days old, with replies unchecked (15 of 25). No changelog or release notes found; renamed settings keep their old names as aliases and a state migration is described in docs/PLUGIN.md (3 of 15). Version 0.1.0 (0). 46 in all.", "checked": "2026-10-30" }, { "key": "performance", "name": "Performance", "weight": 10, "effectiveWeight": 0, "pending": true, "points": 0, "reason": "Pending. Latency is measured per call by our probes, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until the first probe window closes." }, { "key": "schema", "name": "Schema \u0026 documentation", "weight": 13, "effectiveWeight": 16.25, "score": 66, "points": 10.73, "reason": "Framework reading, docs a model can follow, as superpowers was read. JSON plugin manifests for Claude Code, Codex and Copilot, a hooks.json per agent, slash-command files and a settings table with names, defaults and environment variables, and `--json` on two `te` commands, but no schema for its state or output (12 of 25). README, plugin/README.md, docs/PLUGIN.md and docs/ARCHITECTURE.md are Markdown in the repository; no llms.txt (10). The README says what each mechanism does and when it acts, and docs/PLUGIN.md has a known-limitations section and says when to keep micro-skills off (17 of 20). Mechanisms take on, off, propose or envfacts, numeric settings have stated defaults and minimums, and the config code checks choices (10 of 15). Command examples per agent, a troubleshooting section, `te doctor` and an errors.log, with few documented error outputs (12 of 15). A version number in every manifest but no public changelog (5 of 15). 66 in all.", "checked": "2026-10-30" }, { "key": "ergonomics", "name": "Agent ergonomics", "weight": 13, "effectiveWeight": 16.25, "score": 76, "points": 12.35, "reason": "Framework reading, how much a coding agent has to carry. The plugin's job is to shrink context. It adds up to four facts from failures (the proposer's facts are not capped) and one-line notices, while the loop guard's judge and the proposer are extra model calls, each Codex side call carrying Codex's base prompt of about 25,000 tokens by the code's own note (18 of 25). Each mechanism has its own switch and settings, a propose mode for tool pruning and per-session environment variables (15 of 20). Hooks fail open to `{}`, failures go to errors.log and `te doctor`, and the session start says why a deny list stayed a proposal (15 of 20). Writes are atomic under a file lock, the deny-list writer touches only its own entries and refuses symlinks out of the repository, and pruned tools come back with one command (18 of 20). Sensible defaults (micro-skills off, pruning and the guard on); a Python package with TypeScript extensions for Pi and opencode, not SDKs in two languages (10 of 15). 76 in all.", "checked": "2026-10-30" }, { "key": "security", "name": "Security \u0026 auth", "weight": 14, "effectiveWeight": 17.5, "score": 71, "points": 12.43, "reason": "Harness reading of the framework checklist, as superpowers and agent-zero were read. 30 for what leaves the machine and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No credentials of its own and no network code in the package; model calls go through the person's agent CLI with digests (command templates, file basenames, error lines with paths masked) and never file contents, on by default for the loop guard. Installs run from a clone of main, and Codex's install asks the person to trust all hooks, which then run on every tool call (22). Side calls run with tools off (`claude -p --tools \"\"`, `codex exec -s read-only`, Copilot with every tool denied). It never stops a turn, and `toolprune=propose` holds deny lists for approval, but the default `auto` writes deny rules into .claude/settings.local.json after an idle hour (13). docs/PLUGIN.md says learned facts derive from command output a repository controls, keeps micro-skills off by default and says to keep it off or use envfacts in untrusted repositories (11). events.jsonl records every injection, redirect and notice, ledger.jsonl prices each session, and each action is announced in one line (13). Microsoft's standard SECURITY.md points to MSRC, and Microsoft publishes a bounty policy. microsoft.com's security.txt expired on 23 September 2026, and advisories couldn't be checked (12). 71 in all.", "checked": "2026-10-30" }, { "key": "payments", "name": "Payments \u0026 pricing", "weight": 10, "effectiveWeight": 12.5, "score": 60, "points": 7.5, "reason": "Self-hosted rule. No payment protocol (0). Free under Apache-2.0 with nothing to buy (20). No card or account (20). An agent can be given it with no sign-up, through its own plugin manager (20). Its judge and proposer calls are billed on the person's agent plan, and `te status` reports that cost. 60 in all.", "checked": "2026-10-30" }, { "key": "tasks", "name": "Task success", "weight": 10, "effectiveWeight": 0, "pending": true, "points": 0, "reason": "Pending. Task success needs the category task suites run through each tool, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until then. A data provider's data-quality score is published on its listing now and becomes half of this category when it's scored." }, { "key": "maintenance", "name": "Maintenance \u0026 community", "weight": 7, "effectiveWeight": 8.75, "score": 43, "points": 3.76, "reason": "No GitHub release found, and the releases page couldn't be read. The repository was public by 25 October 2026 and docs/PLUGIN.md records verification runs up to 26 October, so the last published change is within 30 days (30). No three releases or dated changelog entries in 90 days (0). Issue and pull-request replies couldn't be read; the archived page shows 1 open issue and 5 open pull requests (5 of 25). Not an MCP server and in no official plugin marketplace (0). Standard-library only with no runtime dependencies, CI on push and pull requests, and the Codex version pinned in CI on purpose (8 of 10). 43 in all.", "checked": "2026-10-30" }, { "key": "transparency", "name": "Transparency \u0026 trust", "weight": 7, "effectiveWeight": 8.75, "score": 76, "points": 6.65, "note": "editorial 84, provenance 67", "reason": "Apache-2.0, an OSI licence (30). docs/PLUGIN.md section 4 and the README agree that learned state stays local and name which model calls leave the machine and what they carry, and the package source has no network code of its own. No privacy statement is linked from the repository (24 of 30). Old setting names and environment variables are still read, and a migration of state keys after 0.1.0 is documented, but there's no written deprecation policy (10 of 20). No telemetry found in the source or docs, and the outbound model calls are disclosed with switches to turn each off (20). 84 in all.", "checked": "2026-10-30" } ], "assessment": { "date": "2026-10-30", "basis": "public evidence", "confidence": "medium", "notes": { "ergonomics": "Framework reading, how much a coding agent has to carry. The plugin's job is to shrink context. It adds up to four facts from failures (the proposer's facts are not capped) and one-line notices, while the loop guard's judge and the proposer are extra model calls, each Codex side call carrying Codex's base prompt of about 25,000 tokens by the code's own note (18 of 25). Each mechanism has its own switch and settings, a propose mode for tool pruning and per-session environment variables (15 of 20). Hooks fail open to `{}`, failures go to errors.log and `te doctor`, and the session start says why a deny list stayed a proposal (15 of 20). Writes are atomic under a file lock, the deny-list writer touches only its own entries and refuses symlinks out of the repository, and pruned tools come back with one command (18 of 20). Sensible defaults (micro-skills off, pruning and the guard on); a Python package with TypeScript extensions for Pi and opencode, not SDKs in two languages (10 of 15). 76 in all.", "maintenance": "No GitHub release found, and the releases page couldn't be read. The repository was public by 25 October 2026 and docs/PLUGIN.md records verification runs up to 26 October, so the last published change is within 30 days (30). No three releases or dated changelog entries in 90 days (0). Issue and pull-request replies couldn't be read; the archived page shows 1 open issue and 5 open pull requests (5 of 25). Not an MCP server and in no official plugin marketplace (0). Standard-library only with no runtime dependencies, CI on push and pull requests, and the Codex version pinned in CI on purpose (8 of 10). 43 in all.", "payments": "Self-hosted rule. No payment protocol (0). Free under Apache-2.0 with nothing to buy (20). No card or account (20). An agent can be given it with no sign-up, through its own plugin manager (20). Its judge and proposer calls are billed on the person's agent plan, and `te status` reports that cost. 60 in all.", "reliability": "Local-package reading, as for superpowers and impeccable. Installs from a git clone through each agent's own plugin manager (a local marketplace on Claude Code, Codex and Copilot, `pi install git:` on Pi), not from a registry or an official marketplace, with Python 3.10 or later and a supported-versions table per agent (14 of 20). .github/workflows/tests.yml runs pytest on push to main and on pull requests, on Ubuntu and Windows with Python 3.10 and 3.13, plus a Codex end-to-end job pinned to 0.164.0. 17 test files in tests/. Whether the runs pass wasn't checked, as github.com couldn't be read (14 of 25). The archived repository page of 30 October 2026 shows 1 open issue and 5 open pull requests, few for a project days old, with replies unchecked (15 of 25). No changelog or release notes found; renamed settings keep their old names as aliases and a state migration is described in docs/PLUGIN.md (3 of 15). Version 0.1.0 (0). 46 in all.", "schema": "Framework reading, docs a model can follow, as superpowers was read. JSON plugin manifests for Claude Code, Codex and Copilot, a hooks.json per agent, slash-command files and a settings table with names, defaults and environment variables, and `--json` on two `te` commands, but no schema for its state or output (12 of 25). README, plugin/README.md, docs/PLUGIN.md and docs/ARCHITECTURE.md are Markdown in the repository; no llms.txt (10). The README says what each mechanism does and when it acts, and docs/PLUGIN.md has a known-limitations section and says when to keep micro-skills off (17 of 20). Mechanisms take on, off, propose or envfacts, numeric settings have stated defaults and minimums, and the config code checks choices (10 of 15). Command examples per agent, a troubleshooting section, `te doctor` and an errors.log, with few documented error outputs (12 of 15). A version number in every manifest but no public changelog (5 of 15). 66 in all.", "security": "Harness reading of the framework checklist, as superpowers and agent-zero were read. 30 for what leaves the machine and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No credentials of its own and no network code in the package; model calls go through the person's agent CLI with digests (command templates, file basenames, error lines with paths masked) and never file contents, on by default for the loop guard. Installs run from a clone of main, and Codex's install asks the person to trust all hooks, which then run on every tool call (22). Side calls run with tools off (`claude -p --tools \"\"`, `codex exec -s read-only`, Copilot with every tool denied). It never stops a turn, and `toolprune=propose` holds deny lists for approval, but the default `auto` writes deny rules into .claude/settings.local.json after an idle hour (13). docs/PLUGIN.md says learned facts derive from command output a repository controls, keeps micro-skills off by default and says to keep it off or use envfacts in untrusted repositories (11). events.jsonl records every injection, redirect and notice, ledger.jsonl prices each session, and each action is announced in one line (13). Microsoft's standard SECURITY.md points to MSRC, and Microsoft publishes a bounty policy. microsoft.com's security.txt expired on 23 September 2026, and advisories couldn't be checked (12). 71 in all.", "transparency": "Apache-2.0, an OSI licence (30). docs/PLUGIN.md section 4 and the README agree that learned state stays local and name which model calls leave the machine and what they carry, and the package source has no network code of its own. No privacy statement is linked from the repository (24 of 30). Old setting names and environment variables are still read, and a migration of state keys after 0.1.0 is documented, but there's no written deprecation policy (10 of 20). No telemetry found in the source or docs, and the outbound model calls are disclosed with switches to turn each off (20). 84 in all." }, "sources": [ { "what": "README", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/README.md", "seen": "2026-10-30" }, { "what": "plugin install and usage guide", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/README.md", "seen": "2026-10-30" }, { "what": "plugin design, verified agent behaviour, state and privacy, known limitations", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/docs/PLUGIN.md", "seen": "2026-10-30" }, { "what": "architecture", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/docs/ARCHITECTURE.md", "seen": "2026-10-30" }, { "what": "contributor conventions", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/CLAUDE.md", "seen": "2026-10-30" }, { "what": "LICENSE (Apache 2.0)", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/LICENSE", "seen": "2026-10-30" }, { "what": "SECURITY.md (Microsoft standard block)", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/SECURITY.md", "seen": "2026-10-30" }, { "what": "pyproject.toml (version 0.1.0, no dependencies)", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/pyproject.toml", "seen": "2026-10-30" }, { "what": "CI workflow", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/.github/workflows/tests.yml", "seen": "2026-10-30" }, { "what": "Claude Code hooks", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/hooks/hooks.json", "seen": "2026-10-30" }, { "what": "hook launcher", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/scripts/te-hook", "seen": "2026-10-30" }, { "what": "Claude Code plugin manifest", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/.claude-plugin/plugin.json", "seen": "2026-10-30" }, { "what": "marketplace manifest", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/.claude-plugin/marketplace.json", "seen": "2026-10-30" }, { "what": "side calls through the agent CLI", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/tokeneconomizer/plugin/sidecall.py", "seen": "2026-10-30" }, { "what": "settings and defaults", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/tokeneconomizer/plugin/config.py", "seen": "2026-10-30" }, { "what": "state store", "url": "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/tokeneconomizer/plugin/store.py", "seen": "2026-10-30" }, { "what": "arXiv abstract, v1 submitted 15 October 2026", "url": "https://arxiv.org/abs/2610.22732", "seen": "2026-10-30" }, { "what": "arXiv HTML v1 (no repository link)", "url": "https://arxiv.org/html/2610.22732v1", "seen": "2026-10-30" }, { "what": "project page", "url": "https://tokeneconomizer.github.io/", "seen": "2026-10-30" }, { "what": "archived GitHub repository page (52 commits, 976 stars, 1 open issue, 5 open PRs, top-level files)", "url": "https://web.archive.org/web/20261030105749/https://github.com/microsoft/TokenEconomizer", "seen": "2026-10-30" }, { "what": "microsoft.com security.txt (Expires 2026-09-23)", "url": "https://www.microsoft.com/.well-known/security.txt", "seen": "2026-10-30" }, { "what": "EveryDev listing, datePublished 2026-10-25 (third party, used only for the date)", "url": "https://www.everydev.ai/tools/tokeneconomizer", "seen": "2026-10-30" } ], "openQuestions": [ "firstReleased is 2026-10-25, the datePublished of EveryDev's listing that links the public repository; the repository's creation date and first commit couldn't be read (github.com and api.github.com unreachable from this session). The arXiv paper v1 of 15 October 2026 has no repository link. The README's X post link carries a post id that decodes to 26 October 2026; X wasn't read. The repository's own docs name verification runs up to 26 October 2026", "unchecked: GitHub created_at, tags, releases, contributors, issue and pull-request replies, Actions results and private vulnerability reporting; github.com and api.github.com couldn't be read, so counts come from the Wayback Machine's copy of the repository page of 30 October 2026, 10:57 UTC", "unchecked: whether a tokeneconomizer package exists on PyPI. pypi.org answered with a client challenge (a bot check), which we didn't pass. The README installs only from git", "data.jsdelivr.com answered robots.txt with HTTP 400, which the sweep rules treat as blocked. Two GETs were sent to it (its file listing for the repository) before that rule was applied. Its listing was stale (no .github or plugin/opencode) and nothing from it is used as evidence; files were read from raw.githubusercontent.com", "Vendor's claims. 18.4 to 30.9 per cent lower mean cost across five agent and model combinations, cheaper in 30 of 30 cells, Pass@1 up in 4 of 5, and 6.4 to 21.2 per cent on other benchmarks are the paper's figures. No independent measurement was found. No 'first', 'best' or '#1' claim was found on the README or project page, so no deduction", "Judgement call. Listed in code, as the batch suggested and as superpowers is, with code.repo because it learns facts about a repository. No category covers cost control for coding agents; agent-memory (where context-mode is) fits the micro-skills part only. Graded as a framework, the kind superpowers has as another coding-agent plugin", "Disclosure. These grades are written by agents on Anthropic's Claude models, and Claude Code is one of the agents TokenEconomizer supports and was measured on. The checklist was applied as for any listing", "Our own products. TokenEconomizer lowers a coding agent's own token cost. It does neither LocalGhost's job nor letme's (choosing and paying for a tool for a task), so it was graded the normal way", "We did not install or run TokenEconomizer, its hooks or its experiments. Every claim about behaviour comes from the source and docs, read as text" ] }, "negative": 0, "verdict": "A standard-library Python plugin that keeps learned state on the machine, documents each hook and makes every action reversible. It is version 0.1.0 with no changelog, installs from a git clone of main, and the 18.4 to 30.9 per cent savings are the paper's SWE-Bench figures, measured with settings that differ from the plugin's defaults.", "bestFor": "Lowering the repeat cost of coding agents in repositories they work in often", "strengths": [ "No API key or credentials of its own; its model calls go through the agent CLI the person is already signed in to, on the same plan", "Learned state stays under ~/.tokeneconomizer with 0700 and 0600 permissions, and only masked digests of commands and errors go to a model", "Every action can be undone. `toolprune restore` returns pruned tools, contradicted facts are dropped, and the loop guard never stops a turn itself", "Public CI runs the test suite on Ubuntu and Windows with Python 3.10 and 3.13, plus a Codex CLI end-to-end job pinned to 0.164.0", "Apache-2.0 and standard-library only, with docs/PLUGIN.md recording how each agent's hooks and settings were verified to behave" ], "weaknesses": [ "Version 0.1.0 with no changelog or release notes found; updating means a `git pull` of main", "Not in any official plugin marketplace. Claude Code, Codex and Copilot install from a local clone, and Codex asks the person to trust all hooks", "Supported agent versions are narrow (Claude Code 2.1.300 to 2.1.310, Codex 0.153.1 to 0.164.0), and newer versions are untested", "The headline savings come from the paper's runs with micro-skills on and side calls on the agent's own model; the plugin defaults micro-skills off and uses Haiku on Claude Code", "Learned facts derive from command output a repository controls, and the proposer's facts are free text; the docs say to keep micro-skills off in untrusted repositories" ], "agentNotes": [ "Check the agent's version against the README's supported range before installing; versions outside it are untested or unsupported.", "Keep micro-skills off, or set microskills=envfacts, in a repository whose command output you don't trust.", "If a tool you need is missing, run the `toolprune` command to see what was left out and `toolprune restore` to bring it back.", "Run the plugin's commands from a terminal with ~/.tokeneconomizer/bin/te; inside the agent each command costs a model turn.", "Start the agent with TE_DISABLE=1 to run one session without the plugin." ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "medium", "grade": "B", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 62.6 } ], "editorialScores": { "ergonomics": 76, "maintenance": 43, "payments": 60, "reliability": 46, "schema": 66, "security": 71, "transparency": 84 }, "provenanceScore": 67, "checked": { "categories": { "ergonomics": "2026-10-30", "maintenance": "2026-10-30", "payments": "2026-10-30", "reliability": "2026-10-30", "schema": "2026-10-30", "security": "2026-10-30", "transparency": "2026-10-30" }, "last": "2026-10-30", "oldest": "2026-10-30", "stale": false, "staleAfterDays": 90 }, "scoreHistory": [ { "date": "2026-10-30", "grade": "B", "kind": "listed", "method": "0.7", "score": 62.6 } ] }, "connect": { "install": "git clone https://github.com/microsoft/TokenEconomizer.git \u0026\u0026 claude plugin marketplace add \"$PWD/TokenEconomizer\"\n# Pi:\npi install git:github.com/microsoft/TokenEconomizer", "claudeCode": "claude plugin install tokeneconomizer@tokeneconomizer" }, "letme": { "capability": "https://letme.dev/code.repo", "tool": "https://letme.dev/tokeneconomizer" }, "sameCompany": [ "azure-foundry-fine-tuning", "azure-ai-content-safety", "azure-speech-to-text", "azure-text-to-speech", "microsoft-agent-framework", "microsoft-execution-containers", "microsoft-entra-agent-id", "azure-key-vault", "azure-document-intelligence", "azure-devops-mcp", "microsoft-learn-mcp", "playwright-mcp", "azure-mcp", "azure-maps", "azure-translator", "microsoft-graph-calendar", "azure-blob-storage", "onedrive-sharepoint", "microsoft-teams", "dynamics-365-sales", "copilot-studio-mcp", "power-automate", "foundry-local", "microsoft-decision-1", "microsoft-advertising-api", "microsoft-excel-graph", "outlook-mail-graph", "microsoft-graph-people", "mage-vl-cowork" ], "notable": [ "The paper, arXiv 2610.22732 v1, was submitted on 15 October 2026 by Aaron Li; its HTML version doesn't link the repository (https://arxiv.org/abs/2610.22732)", "The paper reports 18.4 to 30.9 per cent lower mean cost on 277 SWE-Bench tasks across five agent and model combinations, the authors' figures (https://github.com/microsoft/TokenEconomizer#results)", "Nothing leaves the machine but the loop guard's and proposer's model calls, sent through the person's own agent CLI as masked digests, never file contents (https://github.com/microsoft/TokenEconomizer/blob/main/docs/PLUGIN.md)", "docs/PLUGIN.md says learned facts derive from command output a repository controls and to keep micro-skills off or on envfacts in untrusted repositories (https://github.com/microsoft/TokenEconomizer/blob/main/docs/PLUGIN.md)", "CI runs pytest on Ubuntu and Windows with Python 3.10 and 3.13, and a Codex CLI end-to-end job pinned to 0.164.0 (https://github.com/microsoft/TokenEconomizer/blob/main/.github/workflows/tests.yml)", "The Wayback Machine's copy of the repository page of 30 October 2026 shows 52 commits, 976 stars, 1 open issue and 5 open pull requests (https://web.archive.org/web/20261030105749/https://github.com/microsoft/TokenEconomizer)" ], "area": "developer", "details": [ { "label": "Surface listed", "value": "The microsoft/TokenEconomizer repository at main (version 0.1.0), read from raw.githubusercontent.com, the Wayback Machine's copy of its GitHub page, the project page at tokeneconomizer.github.io and arXiv 2610.22732 on 30 October 2026. Nothing was installed or run" }, { "label": "How an agent reaches it", "value": "As a plugin inside the coding agent, through its hooks (Claude Code, Codex, Copilot CLI), an extension (Pi) or a plugin file (opencode, experimental). Commands run as slash commands or from a terminal with `te`. An outside agent can't call it over a network" }, { "label": "Mechanisms and defaults", "value": "Tool pruning on (a tool unused over 10 prompts across 2 sessions, or 5 sessions, is denied after an idle hour). Loop guard on (a judge at 0.7 confidence, one redirect per turn, then a notice). Micro-skills off (facts confirmed by 3 sessions, a proposer every 3 sessions)" }, { "label": "Install routes", "value": "Claude Code, Codex and Copilot CLI from a local clone added as a marketplace; Pi with `pi install git:github.com/microsoft/TokenEconomizer`; opencode by linking a TypeScript file. Supported versions are Claude Code 2.1.300 to 2.1.310, Codex 0.153.1 to 0.164.0, Copilot CLI 1.0.83 to 1.0.102, Pi 1.2.0 to 1.3.0" }, { "label": "What runs on the machine", "value": "A shell launcher starts a standard-library Python dispatcher on SessionStart, UserPromptSubmit, PostToolUse, PostToolUseFailure, Stop and SessionEnd. The judge runs as a detached `te judge` job so the hook returns at once. Tool pruning edits the agent's local settings file and adds it to .git/info/exclude" }, { "label": "Data leaving the machine", "value": "Only the model calls of the loop guard (on by default), the micro-skills proposer (off by default) and `toolprune measure` when run, sent through the person's own agent CLI as digests with paths and numbers masked. State lives under ~/.tokeneconomizer with 0700 and 0600 permissions" }, { "label": "Reported results", "value": "The paper reports 18.4 to 30.9 per cent lower mean cost on 277 SWE-Bench tasks from six repositories across five agent and model combinations, and 6.4 to 21.2 per cent on Web-Bench and two security-auditing benchmarks. These are the authors' figures; experiments/ holds the harness to rerun them" }, { "label": "Company", "value": "The repository is under Microsoft's GitHub organisation with Microsoft's SECURITY.md. The 11 authors are from UC Berkeley, Microsoft Research and Microsoft Security. The project page is hosted from a separate TokenEconomizer GitHub organisation and links back to the repository" }, { "label": "Category", "value": "Listed in code, the closest category, as superpowers is. Its capability is code.repo because it learns and injects facts about a repository; no category covers cost control for coding agents" } ], "provenance": { "legalEntity": "Microsoft Corporation (repository in the microsoft GitHub organisation with Microsoft's SECURITY.md); co-authors at UC Berkeley", "domain": "microsoft.com", "domainRegistered": "1991-05-02", "endpointOnVendorDomain": null, "terms": "https://github.com/microsoft/TokenEconomizer/blob/main/LICENSE", "privacy": "", "statusPage": "", "changelog": "", "securityTxt": "expired", "checked": "2026-10-30", "notes": [ "The README, the project page and arXiv 2610.22732 name the same 11 authors from UC Berkeley, Microsoft Research and Microsoft Security. The plugin manifests name \"TokenEconomizer authors\" and the repository https://github.com/microsoft/TokenEconomizer.", "domainRegistered for microsoft.com is from the other Microsoft listings (power-automate, azure-mcp) and wasn't re-read.", "www.microsoft.com/.well-known/security.txt, read on 30 October 2026, names MSRC as contact and has Expires 2026-09-23T16:00:00.000Z.", "No terms or privacy statement for the plugin. The Apache 2.0 licence stands in for terms, and docs/PLUGIN.md section 4 describes state and what leaves the machine.", "Local software, so there is no endpoint or status page.", "firstReleased is 25 October 2026, from EveryDev's listing (datePublished 2026-10-25T21:00Z) of the public repository. The arXiv v1 of 15 October 2026 has no repository link, and GitHub's creation date couldn't be read." ], "score": 67, "checks": [ { "check": "Legal entity named", "value": "Microsoft Corporation (repository in the microsoft GitHub organisation with Microsoft's SECURITY.md); co-authors at UC Berkeley", "points": 20, "max": 20, "state": "ok" }, { "check": "Domain age", "value": "microsoft.com, registered 1991-05-02 (35 years)", "points": 15, "max": 15, "state": "ok" }, { "check": "Endpoint on the vendor's domain", "value": "no hosted endpoint", "points": 0, "max": 0, "state": "na" }, { "check": "Terms of service", "value": "read, states 2 of the 7 things a reader expects", "points": 5.7, "max": 10, "state": "part" }, { "check": "Privacy policy", "value": "nothing hosted, not scored", "points": 0, "max": 0, "state": "na" }, { "check": "Status page", "value": "not found", "points": 0, "max": 10, "state": "no" }, { "check": "Changelog", "value": "not found", "points": 0, "max": 10, "state": "no" }, { "check": "security.txt", "value": "published but past its Expires date", "points": 5, "max": 10, "state": "part" }, { "check": "People named", "value": "15 named with their role, 1 with a public profile", "points": 10, "max": 10, "state": "ok" }, { "check": "Support route", "value": "help pages", "points": 5, "max": 5, "state": "ok" } ], "policies": [ { "kind": "terms", "url": "https://github.com/microsoft/TokenEconomizer/blob/main/LICENSE", "state": "read", "readAt": "2026-10-30", "words": 1581, "points": 5.7, "max": 10, "expected": [ { "key": "terms.date", "label": "Gives the date it was last updated", "found": false }, { "key": "terms.law", "label": "Names the governing law or courts", "found": false }, { "key": "terms.liability", "label": "States a limit on its liability", "found": true, "quote": "…arising as a result of this License or out of the use or inability to use the Work (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages.", "says": "Rules out indirect and consequential losses, with no cap named in this sentence" }, { "key": "terms.termination", "label": "Says how the agreement or account can be ended", "found": false }, { "key": "terms.changes", "label": "Says how changes to the terms are announced", "found": false }, { "key": "terms.use", "label": "Lists what users may not do", "found": true, "quote": "you may not use this file except in compliance with the License." }, { "key": "terms.sla", "label": "Refers to a service level or uptime commitment", "found": false } ] }, { "kind": "privacy", "url": "", "state": "none-found", "points": 0, "max": 10 } ] }, "humans": { "slug": "tokeneconomizer", "checked": "2026-10-30", "company": "Microsoft", "people": [ { "name": "Bill Gates", "role": "Co-founder and technical adviser, Microsoft", "founder": true, "linkedin": "https://www.linkedin.com/in/williamhgates", "x": "https://x.com/BillGates", "bio": "Bill Gates co-founded Microsoft with Paul Allen in 1975 and led it through its growth in business and personal software. In 2008 he moved to full-time work at the Gates Foundation, which he chairs, and he founded Breakthrough Energy to back clean-technology companies.", "bioSources": [ "https://www.gatesfoundation.org/about/leadership/bill-gates" ], "page": "/people/bill-gates/" }, { "name": "Paul Allen", "role": "Co-founder, Microsoft, left day-to-day work there in 1983", "founder": true, "bio": "Paul Allen co-founded Microsoft with Bill Gates in 1975. He left day-to-day work at the company in 1983 but stayed on its board as vice-chairman, and co-founded Vulcan Inc. in 1986.", "bioSources": [ "https://en.wikipedia.org/wiki/Paul_Allen" ], "page": "/people/paul-allen/" }, { "name": "Satya Nadella", "role": "Chairman and CEO, Microsoft", "photo": "/assets/img/people/microsoft-satya-nadella.webp", "photoSource": "https://news.microsoft.com/source/leadership/", "bio": "Satya Nadella is chairman and chief executive of Microsoft, which he joined in 1992 and has led as CEO since February 2014. He earlier ran its Cloud and Enterprise group and led research and development for its Online Services Division, and before Microsoft was on the technology staff at Sun Microsystems.", "bioSources": [ "https://news.microsoft.com/source/exec/satya-nadella/" ], "page": "/people/satya-nadella/" }, { "name": "Kevin Scott", "role": "Chief Technology Officer, Microsoft", "photo": "/assets/img/people/microsoft-kevin-scott.webp", "photoSource": "https://news.microsoft.com/source/exec/kevin-scott/", "bio": "Kevin Scott is Microsoft's chief technology officer. He was earlier senior vice president of engineering and operations at LinkedIn, which he led through its IPO, and before that oversaw mobile ads engineering at Google, including the integration of AdMob. He hosts the Behind the Tech podcast and sits on Shopify's board.", "bioSources": [ "https://news.microsoft.com/source/exec/kevin-scott/" ], "page": "/people/kevin-scott/" }, { "name": "Aaron J. Li", "role": "First author of TokenEconomizer, UC Berkeley", "page": "/people/aaron-j-li/" }, { "name": "Weijia Xu", "role": "Co-author of TokenEconomizer, Microsoft Research", "page": "/people/weijia-xu/" }, { "name": "Chandan Singh", "role": "Co-author of TokenEconomizer, Microsoft Research", "page": "/people/chandan-singh/" }, { "name": "Zixiao Chen", "role": "Co-author of TokenEconomizer, Microsoft Security", "page": "/people/zixiao-chen/" }, { "name": "Jeevana Priya Inala", "role": "Co-author of TokenEconomizer, Microsoft Research", "page": "/people/jeevana-priya-inala/" }, { "name": "Charlotte Siska", "role": "Co-author of TokenEconomizer, Microsoft Security", "page": "/people/charlotte-siska/" }, { "name": "Acar Tamersoy", "role": "Co-author of TokenEconomizer, Microsoft Security", "page": "/people/acar-tamersoy/" }, { "name": "Bin Yu", "role": "Co-author of TokenEconomizer, UC Berkeley", "page": "/people/bin-yu/" }, { "name": "Ion Stoica", "role": "Co-author of TokenEconomizer, UC Berkeley", "page": "/people/ion-stoica/" }, { "name": "Michel Galley", "role": "Co-author of TokenEconomizer, Microsoft Research", "page": "/people/michel-galley/" }, { "name": "Jianfeng Gao", "role": "Co-author of TokenEconomizer, Microsoft Research", "page": "/people/jianfeng-gao/" } ], "peopleNote": "The 11 authors and their affiliations (UC Berkeley, Microsoft Research, Microsoft Security) are from the README, the project page and arXiv 2610.22732, read on 30 October 2026, in the paper's order. No author's LinkedIn, X or GitHub account is linked from those pages by name, so none is listed; the README links an X post under the handle aaron_jli, which wasn't read. Microsoft's founders and leadership are as on the other Microsoft listings (azure-mcp, checked 28 October 2026). Microsoft is a public company, so no investors are listed.", "companySocial": { "linkedin": "https://www.linkedin.com/company/microsoft", "x": "https://x.com/Microsoft", "github": "https://github.com/microsoft" }, "support": { "urls": [ "https://github.com/microsoft/TokenEconomizer/issues" ], "contactUrls": [ "https://msrc.microsoft.com/report/vulnerability/new" ] }, "sources": [ "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/README.md", "https://tokeneconomizer.github.io/", "https://arxiv.org/abs/2610.22732", "https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/SECURITY.md", "https://www.microsoft.com/.well-known/security.txt", "https://web.archive.org/web/20261030105749/https://github.com/microsoft/TokenEconomizer", "https://www.gatesfoundation.org/about/leadership/bill-gates", "https://en.wikipedia.org/wiki/Paul_Allen", "https://news.microsoft.com/source/exec/satya-nadella/", "https://news.microsoft.com/source/exec/kevin-scott/" ], "ownership": { "status": "public", "exchange": "NASDAQ", "ticker": "MSFT", "checked": "2026-10-28", "sources": [ "https://www.nasdaq.com/market-activity/stocks/msft", "https://en.wikipedia.org/wiki/Microsoft" ] } }, "pageJsonUrl": "https://www.anchorterminal.com/tools/tokeneconomizer.json" }, "verify": { "accepts": "a page on microsoft.com or tokeneconomizer.github.io or one of their subdomains, or the README of github.com/microsoft/TokenEconomizer", "badgeUrl": "https://www.anchorterminal.com/badges/tokeneconomizer.svg", "body": { "slug": "tokeneconomizer", "url": "the page with the badge or the link" }, "docs": "https://www.anchorterminal.com/builders/#verify", "effect": "none, it never changes a grade, rank or review", "endpoint": "https://www.anchorterminal.com/api/v1/verify", "listingUrl": "https://www.anchorterminal.com/tools/tokeneconomizer", "mcpTool": "verify_listing", "recheck": "weekly; two failed checks in a row and it lapses, a later pass restores it", "snippets": { "html": "\u003ca href=\"https://www.anchorterminal.com/tools/tokeneconomizer\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/tokeneconomizer.svg\" alt=\"TokenEconomizer on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e", "markdown": "[![TokenEconomizer on Anchor Terminal](https://www.anchorterminal.com/badges/tokeneconomizer.svg)](https://www.anchorterminal.com/tools/tokeneconomizer)", "link": "\u003ca href=\"https://www.anchorterminal.com/tools/tokeneconomizer\"\u003eTokenEconomizer on Anchor Terminal\u003c/a\u003e" } } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/tools/tokeneconomizer", "json": "https://www.anchorterminal.com/tools/tokeneconomizer.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/tools/tokeneconomizer.md", "mcp": "https://www.anchorterminal.com/mcp", "pricing": "https://www.anchorterminal.com/tools/tokeneconomizer#pricing", "reviews": "https://www.anchorterminal.com/tools/tokeneconomizer#reviews", "slim": "https://www.anchorterminal.com/tools/tokeneconomizer.min.md" }, "markdown": "## Overview\n\n**Grade B · 62.6/100 · rank #502 of 1244 · #9 in Code \u0026 developer platforms · not agent-ready · confidence medium**\n\n\nMore from Microsoft, listed separately because each is its own product: [Microsoft Foundry fine-tuning (Azure OpenAI)](https://www.anchorterminal.com/tools/azure-foundry-fine-tuning.md) (Fine-tuning), [Azure AI Content Safety (Prompt Shields)](https://www.anchorterminal.com/tools/azure-ai-content-safety.md) (Guardrails \u0026 safety filters), [Azure AI Speech speech-to-text](https://www.anchorterminal.com/tools/azure-speech-to-text.md) (Speech-to-text), [Azure AI Speech text-to-speech](https://www.anchorterminal.com/tools/azure-text-to-speech.md) (Text-to-speech), [Microsoft Agent Framework](https://www.anchorterminal.com/tools/microsoft-agent-framework.md) (Agent frameworks \u0026 SDKs), [Microsoft Execution Containers](https://www.anchorterminal.com/tools/microsoft-execution-containers.md) (Code execution sandboxes), [Microsoft Entra Agent ID](https://www.anchorterminal.com/tools/microsoft-entra-agent-id.md) (Agent auth \u0026 delegated access), [Azure Key Vault](https://www.anchorterminal.com/tools/azure-key-vault.md) (Secrets \u0026 credential vaults), [Azure Document Intelligence](https://www.anchorterminal.com/tools/azure-document-intelligence.md) (Document parsing \u0026 extraction), [Azure DevOps MCP Server](https://www.anchorterminal.com/tools/azure-devops-mcp.md) (Code \u0026 developer platforms), [Microsoft Learn MCP Server](https://www.anchorterminal.com/tools/microsoft-learn-mcp.md) (Code \u0026 developer platforms), [Playwright MCP](https://www.anchorterminal.com/tools/playwright-mcp.md) (Browser automation), [Azure MCP Server](https://www.anchorterminal.com/tools/azure-mcp.md) (Cloud \u0026 infrastructure), [Azure Maps](https://www.anchorterminal.com/tools/azure-maps.md) (Maps, geocoding \u0026 places), [Azure Translator](https://www.anchorterminal.com/tools/azure-translator.md) (Translation), [Microsoft Graph Calendar API](https://www.anchorterminal.com/tools/microsoft-graph-calendar.md) (Calendars \u0026 scheduling), [Azure Blob Storage](https://www.anchorterminal.com/tools/azure-blob-storage.md) (File storage \u0026 sharing), [OneDrive and SharePoint files (Microsoft Graph)](https://www.anchorterminal.com/tools/onedrive-sharepoint.md) (File storage \u0026 sharing), [Microsoft Teams (Microsoft Graph)](https://www.anchorterminal.com/tools/microsoft-teams.md) (Work \u0026 productivity), [Microsoft Dynamics 365 Sales](https://www.anchorterminal.com/tools/dynamics-365-sales.md) (CRM \u0026 customer platforms), [Microsoft Copilot Studio MCP](https://www.anchorterminal.com/tools/copilot-studio-mcp.md) (Workflow automation), [Microsoft Power Automate](https://www.anchorterminal.com/tools/power-automate.md) (Workflow automation), [Foundry Local](https://www.anchorterminal.com/tools/foundry-local.md) (Local AI), [Microsoft-Decision-1](https://www.anchorterminal.com/tools/microsoft-decision-1.md) (Decision models), [Microsoft Advertising API](https://www.anchorterminal.com/tools/microsoft-advertising-api.md) (Advertising \u0026 campaign operations), [Microsoft Excel (Microsoft Graph workbook API)](https://www.anchorterminal.com/tools/microsoft-excel-graph.md) (Spreadsheets \u0026 operational tables), [Outlook Mail (Microsoft Graph)](https://www.anchorterminal.com/tools/outlook-mail-graph.md) (Mailbox access), [Microsoft Graph People and Contacts API](https://www.anchorterminal.com/tools/microsoft-graph-people.md) (Contacts \u0026 address books), [Mage-VL-Cowork](https://www.anchorterminal.com/tools/mage-vl-cowork.md) (Desktop \u0026 mobile app control).\n\n## Assessment\n\nA standard-library Python plugin that keeps learned state on the machine, documents each hook and makes every action reversible. It is version 0.1.0 with no changelog, installs from a git clone of main, and the 18.4 to 30.9 per cent savings are the paper's SWE-Bench figures, measured with settings that differ from the plugin's defaults.\n\n## Facts\n\n| Field | Value |\n| --- | --- |\n| Vendor | Microsoft (https://tokeneconomizer.github.io/) |\n| Kind | Agent framework |\n| Category | Code \u0026 developer platforms (https://www.anchorterminal.com/categories/code) |\n| Auth | None · No account, key or login of its own. It runs inside Claude Code, Codex, the Copilot CLI or Pi, and its own model calls go through that agent's CLI on whatever plan or login the person already has. |\n| Pricing | Free (Free · OSS) · Free under Apache 2.0, with nothing to buy. Its loop-guard judge, the redirect writer and the optional micro-skills proposer make model calls billed on the person's agent plan, Haiku by default on Claude Code, and `te status` reports that cost. |\n| x402 | No · Local software with nothing to buy. No x402, MPP or L402 in the repository or on the project page, checked 30 October 2026. |\n| Licence | Apache-2.0 |\n| Source | https://github.com/microsoft/TokenEconomizer |\n| Docs | https://github.com/microsoft/TokenEconomizer/blob/main/plugin/README.md |\n| llms.txt | not found |\n| GitHub stars | 976 (as of 2026-10-30) |\n| Surface listed | The microsoft/TokenEconomizer repository at main (version 0.1.0), read from raw.githubusercontent.com, the Wayback Machine's copy of its GitHub page, the project page at tokeneconomizer.github.io and arXiv 2610.22732 on 30 October 2026. Nothing was installed or run |\n| How an agent reaches it | As a plugin inside the coding agent, through its hooks (Claude Code, Codex, Copilot CLI), an extension (Pi) or a plugin file (opencode, experimental). Commands run as slash commands or from a terminal with `te`. An outside agent can't call it over a network |\n| Mechanisms and defaults | Tool pruning on (a tool unused over 10 prompts across 2 sessions, or 5 sessions, is denied after an idle hour). Loop guard on (a judge at 0.7 confidence, one redirect per turn, then a notice). Micro-skills off (facts confirmed by 3 sessions, a proposer every 3 sessions) |\n| Install routes | Claude Code, Codex and Copilot CLI from a local clone added as a marketplace; Pi with `pi install git:github.com/microsoft/TokenEconomizer`; opencode by linking a TypeScript file. Supported versions are Claude Code 2.1.300 to 2.1.310, Codex 0.153.1 to 0.164.0, Copilot CLI 1.0.83 to 1.0.102, Pi 1.2.0 to 1.3.0 |\n| What runs on the machine | A shell launcher starts a standard-library Python dispatcher on SessionStart, UserPromptSubmit, PostToolUse, PostToolUseFailure, Stop and SessionEnd. The judge runs as a detached `te judge` job so the hook returns at once. Tool pruning edits the agent's local settings file and adds it to .git/info/exclude |\n| Data leaving the machine | Only the model calls of the loop guard (on by default), the micro-skills proposer (off by default) and `toolprune measure` when run, sent through the person's own agent CLI as digests with paths and numbers masked. State lives under ~/.tokeneconomizer with 0700 and 0600 permissions |\n| Reported results | The paper reports 18.4 to 30.9 per cent lower mean cost on 277 SWE-Bench tasks from six repositories across five agent and model combinations, and 6.4 to 21.2 per cent on Web-Bench and two security-auditing benchmarks. These are the authors' figures; experiments/ holds the harness to rerun them |\n| Company | The repository is under Microsoft's GitHub organisation with Microsoft's SECURITY.md. The 11 authors are from UC Berkeley, Microsoft Research and Microsoft Security. The project page is hosted from a separate TokenEconomizer GitHub organisation and links back to the repository |\n| Category | Listed in code, the closest category, as superpowers is. Its capability is code.repo because it learns and injects facts about a repository; no category covers cost control for coding agents |\n| Capabilities | code.repo |\n| Tags | open-source, apache-2.0, local, free, python, claude-code-plugin, hooks, pre-1.0 |\n| JSON | https://www.anchorterminal.com/api/v1/tools/tokeneconomizer.json |\n\n## Score breakdown (methodology v0.7, October 2026 research run)\n\nAssessed 2026-10-30 from public evidence against the published checklist (https://www.anchorterminal.com/benchmark/#checklist). Confidence: medium. Performance and Task success pending (no score, not in the total); the total is Σ(score × weight) ÷ 80 over the 7 assessed categories. \"This run\" is each category's share of the 100 points.\n\n| Category | Weight | This run | Score (0–100) | Points |\n| --- | --- | --- | --- | --- |\n| Reliability | 16% | 20 | 46 | 9.2 |\n| Performance | 10% | pending | pending | n/a |\n| Schema \u0026 documentation | 13% | 16.2 | 66 | 10.7 |\n| Agent ergonomics | 13% | 16.2 | 76 | 12.3 |\n| Security \u0026 auth | 14% | 17.5 | 71 | 12.4 |\n| Payments \u0026 pricing | 10% | 12.5 | 60 | 7.5 |\n| Task success | 10% | pending | pending | n/a |\n| Maintenance \u0026 community | 7% | 8.8 | 43 | 3.8 |\n| Transparency \u0026 trust (editorial 84, provenance 67) | 7% | 8.8 | 76 | 6.7 |\n| Negative events | up to −15 | up to −15 | none recorded | 0 |\n| **Total** | | | | **62.6 → B** |\n\n### Why each score\n\n- Reliability 46 (checked 2026-10-30): Local-package reading, as for superpowers and impeccable. Installs from a git clone through each agent's own plugin manager (a local marketplace on Claude Code, Codex and Copilot, `pi install git:` on Pi), not from a registry or an official marketplace, with Python 3.10 or later and a supported-versions table per agent (14 of 20). .github/workflows/tests.yml runs pytest on push to main and on pull requests, on Ubuntu and Windows with Python 3.10 and 3.13, plus a Codex end-to-end job pinned to 0.164.0. 17 test files in tests/. Whether the runs pass wasn't checked, as github.com couldn't be read (14 of 25). The archived repository page of 30 October 2026 shows 1 open issue and 5 open pull requests, few for a project days old, with replies unchecked (15 of 25). No changelog or release notes found; renamed settings keep their old names as aliases and a state migration is described in docs/PLUGIN.md (3 of 15). Version 0.1.0 (0). 46 in all.\n- Performance: Pending. Latency is measured per call by our probes, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until the first probe window closes.\n- Schema \u0026 documentation 66 (checked 2026-10-30): Framework reading, docs a model can follow, as superpowers was read. JSON plugin manifests for Claude Code, Codex and Copilot, a hooks.json per agent, slash-command files and a settings table with names, defaults and environment variables, and `--json` on two `te` commands, but no schema for its state or output (12 of 25). README, plugin/README.md, docs/PLUGIN.md and docs/ARCHITECTURE.md are Markdown in the repository; no llms.txt (10). The README says what each mechanism does and when it acts, and docs/PLUGIN.md has a known-limitations section and says when to keep micro-skills off (17 of 20). Mechanisms take on, off, propose or envfacts, numeric settings have stated defaults and minimums, and the config code checks choices (10 of 15). Command examples per agent, a troubleshooting section, `te doctor` and an errors.log, with few documented error outputs (12 of 15). A version number in every manifest but no public changelog (5 of 15). 66 in all.\n- Agent ergonomics 76 (checked 2026-10-30): Framework reading, how much a coding agent has to carry. The plugin's job is to shrink context. It adds up to four facts from failures (the proposer's facts are not capped) and one-line notices, while the loop guard's judge and the proposer are extra model calls, each Codex side call carrying Codex's base prompt of about 25,000 tokens by the code's own note (18 of 25). Each mechanism has its own switch and settings, a propose mode for tool pruning and per-session environment variables (15 of 20). Hooks fail open to `{}`, failures go to errors.log and `te doctor`, and the session start says why a deny list stayed a proposal (15 of 20). Writes are atomic under a file lock, the deny-list writer touches only its own entries and refuses symlinks out of the repository, and pruned tools come back with one command (18 of 20). Sensible defaults (micro-skills off, pruning and the guard on); a Python package with TypeScript extensions for Pi and opencode, not SDKs in two languages (10 of 15). 76 in all.\n- Security \u0026 auth 71 (checked 2026-10-30): Harness reading of the framework checklist, as superpowers and agent-zero were read. 30 for what leaves the machine and credentials, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. No credentials of its own and no network code in the package; model calls go through the person's agent CLI with digests (command templates, file basenames, error lines with paths masked) and never file contents, on by default for the loop guard. Installs run from a clone of main, and Codex's install asks the person to trust all hooks, which then run on every tool call (22). Side calls run with tools off (`claude -p --tools \"\"`, `codex exec -s read-only`, Copilot with every tool denied). It never stops a turn, and `toolprune=propose` holds deny lists for approval, but the default `auto` writes deny rules into .claude/settings.local.json after an idle hour (13). docs/PLUGIN.md says learned facts derive from command output a repository controls, keeps micro-skills off by default and says to keep it off or use envfacts in untrusted repositories (11). events.jsonl records every injection, redirect and notice, ledger.jsonl prices each session, and each action is announced in one line (13). Microsoft's standard SECURITY.md points to MSRC, and Microsoft publishes a bounty policy. microsoft.com's security.txt expired on 23 September 2026, and advisories couldn't be checked (12). 71 in all.\n- Payments \u0026 pricing 60 (checked 2026-10-30): Self-hosted rule. No payment protocol (0). Free under Apache-2.0 with nothing to buy (20). No card or account (20). An agent can be given it with no sign-up, through its own plugin manager (20). Its judge and proposer calls are billed on the person's agent plan, and `te status` reports that cost. 60 in all.\n- Task success: Pending. Task success needs the category task suites run through each tool, which haven't run yet, so this run doesn't score it. Its weight is shared across the assessed categories until then. A data provider's data-quality score is published on its listing now and becomes half of this category when it's scored.\n- Maintenance \u0026 community 43 (checked 2026-10-30): No GitHub release found, and the releases page couldn't be read. The repository was public by 25 October 2026 and docs/PLUGIN.md records verification runs up to 26 October, so the last published change is within 30 days (30). No three releases or dated changelog entries in 90 days (0). Issue and pull-request replies couldn't be read; the archived page shows 1 open issue and 5 open pull requests (5 of 25). Not an MCP server and in no official plugin marketplace (0). Standard-library only with no runtime dependencies, CI on push and pull requests, and the Codex version pinned in CI on purpose (8 of 10). 43 in all.\n- Transparency \u0026 trust 76 (checked 2026-10-30): Apache-2.0, an OSI licence (30). docs/PLUGIN.md section 4 and the README agree that learned state stays local and name which model calls leave the machine and what they carry, and the package source has no network code of its own. No privacy statement is linked from the repository (24 of 30). Old setting names and environment variables are still read, and a migration of state keys after 0.1.0 is documented, but there's no written deprecation policy (10 of 20). No telemetry found in the source or docs, and the outbound model calls are disclosed with switches to turn each off (20). 84 in all.\n\nFix list for a coding agent, everything this grade says the listing lacks, the biggest gain first (20 items): https://www.anchorterminal.com/fixes/tokeneconomizer.md (JSON https://www.anchorterminal.com/fixes/tokeneconomizer.json)\n\n### What we couldn't check\n\n- firstReleased is 2026-10-25, the datePublished of EveryDev's listing that links the public repository; the repository's creation date and first commit couldn't be read (github.com and api.github.com unreachable from this session). The arXiv paper v1 of 15 October 2026 has no repository link. The README's X post link carries a post id that decodes to 26 October 2026; X wasn't read. The repository's own docs name verification runs up to 26 October 2026\n- unchecked: GitHub created_at, tags, releases, contributors, issue and pull-request replies, Actions results and private vulnerability reporting; github.com and api.github.com couldn't be read, so counts come from the Wayback Machine's copy of the repository page of 30 October 2026, 10:57 UTC\n- unchecked: whether a tokeneconomizer package exists on PyPI. pypi.org answered with a client challenge (a bot check), which we didn't pass. The README installs only from git\n- data.jsdelivr.com answered robots.txt with HTTP 400, which the sweep rules treat as blocked. Two GETs were sent to it (its file listing for the repository) before that rule was applied. Its listing was stale (no .github or plugin/opencode) and nothing from it is used as evidence; files were read from raw.githubusercontent.com\n- Vendor's claims. 18.4 to 30.9 per cent lower mean cost across five agent and model combinations, cheaper in 30 of 30 cells, Pass@1 up in 4 of 5, and 6.4 to 21.2 per cent on other benchmarks are the paper's figures. No independent measurement was found. No 'first', 'best' or '#1' claim was found on the README or project page, so no deduction\n- Judgement call. Listed in code, as the batch suggested and as superpowers is, with code.repo because it learns facts about a repository. No category covers cost control for coding agents; agent-memory (where context-mode is) fits the micro-skills part only. Graded as a framework, the kind superpowers has as another coding-agent plugin\n- Disclosure. These grades are written by agents on Anthropic's Claude models, and Claude Code is one of the agents TokenEconomizer supports and was measured on. The checklist was applied as for any listing\n- Our own products. TokenEconomizer lowers a coding agent's own token cost. It does neither LocalGhost's job nor letme's (choosing and paying for a tool for a task), so it was graded the normal way\n- We did not install or run TokenEconomizer, its hooks or its experiments. Every claim about behaviour comes from the source and docs, read as text\n\n### Sources\n\n- README: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/README.md\u003e (seen 2026-10-30)\n- plugin install and usage guide: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/README.md\u003e (seen 2026-10-30)\n- plugin design, verified agent behaviour, state and privacy, known limitations: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/docs/PLUGIN.md\u003e (seen 2026-10-30)\n- architecture: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/docs/ARCHITECTURE.md\u003e (seen 2026-10-30)\n- contributor conventions: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/CLAUDE.md\u003e (seen 2026-10-30)\n- LICENSE (Apache 2.0): \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/LICENSE\u003e (seen 2026-10-30)\n- SECURITY.md (Microsoft standard block): \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/SECURITY.md\u003e (seen 2026-10-30)\n- pyproject.toml (version 0.1.0, no dependencies): \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/pyproject.toml\u003e (seen 2026-10-30)\n- CI workflow: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/.github/workflows/tests.yml\u003e (seen 2026-10-30)\n- Claude Code hooks: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/hooks/hooks.json\u003e (seen 2026-10-30)\n- hook launcher: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/scripts/te-hook\u003e (seen 2026-10-30)\n- Claude Code plugin manifest: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/plugin/.claude-plugin/plugin.json\u003e (seen 2026-10-30)\n- marketplace manifest: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/.claude-plugin/marketplace.json\u003e (seen 2026-10-30)\n- side calls through the agent CLI: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/tokeneconomizer/plugin/sidecall.py\u003e (seen 2026-10-30)\n- settings and defaults: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/tokeneconomizer/plugin/config.py\u003e (seen 2026-10-30)\n- state store: \u003chttps://raw.githubusercontent.com/microsoft/TokenEconomizer/main/tokeneconomizer/plugin/store.py\u003e (seen 2026-10-30)\n- arXiv abstract, v1 submitted 15 October 2026: \u003chttps://arxiv.org/abs/2610.22732\u003e (seen 2026-10-30)\n- arXiv HTML v1 (no repository link): \u003chttps://arxiv.org/html/2610.22732v1\u003e (seen 2026-10-30)\n- project page: \u003chttps://tokeneconomizer.github.io/\u003e (seen 2026-10-30)\n- archived GitHub repository page (52 commits, 976 stars, 1 open issue, 5 open PRs, top-level files): \u003chttps://web.archive.org/web/20261030105749/https://github.com/microsoft/TokenEconomizer\u003e (seen 2026-10-30)\n- microsoft.com security.txt (Expires 2026-09-23): \u003chttps://www.microsoft.com/.well-known/security.txt\u003e (seen 2026-10-30)\n- EveryDev listing, datePublished 2026-10-25 (third party, used only for the date): \u003chttps://www.everydev.ai/tools/tokeneconomizer\u003e (seen 2026-10-30)\n\n## Score history\n\nEvery move of a score since the listing was graded, newest first, with its cause. A score can't move without an entry here. JSON: https://www.anchorterminal.com/api/v1/tools/tokeneconomizer/history.json\n\n- 2026-10-30: graded 62.6 B (methodology 0.7)\n\n## Who's behind it (provenance 67/100, checked 2026-10-30)\n\n| Check | Finding | Points |\n| --- | --- | --- |\n| Legal entity named | Microsoft Corporation (repository in the microsoft GitHub organisation with Microsoft's SECURITY.md); co-authors at UC Berkeley | 20/20 |\n| Domain age | microsoft.com, registered 1991-05-02 (35 years) | 15/15 |\n| Endpoint on the vendor's domain | no hosted endpoint | n/a |\n| Terms of service | read, states 2 of the 7 things a reader expects | 5.7/10 |\n| Privacy policy | nothing hosted, not scored | n/a |\n| Status page | not found | 0/10 |\n| Changelog | not found | 0/10 |\n| security.txt | published but past its Expires date | 5/10 |\n| People named | 15 named with their role, 1 with a public profile | 10/10 |\n| Support route | help pages | 5/5 |\n\nThe README, the project page and arXiv 2610.22732 name the same 11 authors from UC Berkeley, Microsoft Research and Microsoft Security. The plugin manifests name \"TokenEconomizer authors\" and the repository https://github.com/microsoft/TokenEconomizer.\n\ndomainRegistered for microsoft.com is from the other Microsoft listings (power-automate, azure-mcp) and wasn't re-read.\n\nwww.microsoft.com/.well-known/security.txt, read on 30 October 2026, names MSRC as contact and has Expires 2026-09-23T16:00:00.000Z.\n\nNo terms or privacy statement for the plugin. The Apache 2.0 licence stands in for terms, and docs/PLUGIN.md section 4 describes state and what leaves the machine.\n\nLocal software, so there is no endpoint or status page.\n\nfirstReleased is 25 October 2026, from EveryDev's listing (datePublished 2026-10-25T21:00Z) of the public repository. The arXiv v1 of 15 October 2026 has no repository link, and GitHub's creation date couldn't be read.\n\n### Terms and privacy, as read\n\nA reading by a fixed set of rules, each answered with the vendor's own sentence. Not legal advice.\n\n**Terms of service** (https://github.com/microsoft/TokenEconomizer/blob/main/LICENSE), read 2026-10-30, gives no date, states 2 of the 7 things a reader expects.\n\n- Not found in the text. Gives the date it was last updated.\n- Not found in the text. Names the governing law or courts.\n- States a limit on its liability. Rules out indirect and consequential losses, with no cap named in this sentence.\n- Not found in the text. Says how the agreement or account can be ended.\n- Not found in the text. Says how changes to the terms are announced.\n- Not found in the text. Refers to a service level or uptime commitment.\n\n**Privacy policy**. Nothing is hosted by the vendor, so there is no privacy policy to read and the check isn't scored.\n\n\n## Humans behind the project (researched 2026-10-30 from public pages, not part of the grade)\n\n### Key people\n\n- Bill Gates, Co-founder and technical adviser, Microsoft. LinkedIn: https://www.linkedin.com/in/williamhgates. X: https://x.com/BillGates\n- Paul Allen, Co-founder, Microsoft, left day-to-day work there in 1983\n- Satya Nadella, Chairman and CEO, Microsoft\n- Kevin Scott, Chief Technology Officer, Microsoft\n- Aaron J. Li, First author of TokenEconomizer, UC Berkeley\n- Weijia Xu, Co-author of TokenEconomizer, Microsoft Research\n- Chandan Singh, Co-author of TokenEconomizer, Microsoft Research\n- Zixiao Chen, Co-author of TokenEconomizer, Microsoft Security\n- Jeevana Priya Inala, Co-author of TokenEconomizer, Microsoft Research\n- Charlotte Siska, Co-author of TokenEconomizer, Microsoft Security\n- Acar Tamersoy, Co-author of TokenEconomizer, Microsoft Security\n- Bin Yu, Co-author of TokenEconomizer, UC Berkeley\n- Ion Stoica, Co-author of TokenEconomizer, UC Berkeley\n- Michel Galley, Co-author of TokenEconomizer, Microsoft Research\n- Jianfeng Gao, Co-author of TokenEconomizer, Microsoft Research\n- Note: the 11 authors and their affiliations (uc berkeley, microsoft research, microsoft security) are from the readme, the project page and arxiv 2610.22732, read on 30 october 2026, in the paper's order. no author's linkedin, x or github account is linked from those pages by name, so none is listed; the readme links an x post under the handle aaron_jli, which wasn't read. microsoft's founders and leadership are as on the other microsoft listings (azure-mcp, checked 28 october 2026). microsoft is a public company, so no investors are listed.\n- Microsoft: LinkedIn https://www.linkedin.com/company/microsoft X https://x.com/Microsoft GitHub https://github.com/microsoft\n\n### Ownership and investors\n\n- Ownership: Public company (NASDAQ: MSFT)\n\n### Getting support\n\n- Help: https://github.com/microsoft/TokenEconomizer/issues\n- Sources: https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/README.md, https://tokeneconomizer.github.io/, https://arxiv.org/abs/2610.22732, https://raw.githubusercontent.com/microsoft/TokenEconomizer/main/SECURITY.md, https://www.microsoft.com/.well-known/security.txt, https://web.archive.org/web/20261030105749/https://github.com/microsoft/TokenEconomizer, https://www.gatesfoundation.org/about/leadership/bill-gates, https://en.wikipedia.org/wiki/Paul_Allen, https://news.microsoft.com/source/exec/satya-nadella/, https://news.microsoft.com/source/exec/kevin-scott/\n\n## Probe metrics\n\nNot measured yet. Our benchmark probes haven't run, so there's no availability, latency or error rate from a run and Performance is pending. Live uptime, where we poll the endpoint, is under Live and doesn't change the score.\n\n## Strengths\n\n- No API key or credentials of its own; its model calls go through the agent CLI the person is already signed in to, on the same plan\n- Learned state stays under ~/.tokeneconomizer with 0700 and 0600 permissions, and only masked digests of commands and errors go to a model\n- Every action can be undone. `toolprune restore` returns pruned tools, contradicted facts are dropped, and the loop guard never stops a turn itself\n- Public CI runs the test suite on Ubuntu and Windows with Python 3.10 and 3.13, plus a Codex CLI end-to-end job pinned to 0.164.0\n- Apache-2.0 and standard-library only, with docs/PLUGIN.md recording how each agent's hooks and settings were verified to behave\n\n## Weaknesses\n\n- Version 0.1.0 with no changelog or release notes found; updating means a `git pull` of main\n- Not in any official plugin marketplace. Claude Code, Codex and Copilot install from a local clone, and Codex asks the person to trust all hooks\n- Supported agent versions are narrow (Claude Code 2.1.300 to 2.1.310, Codex 0.153.1 to 0.164.0), and newer versions are untested\n- The headline savings come from the paper's runs with micro-skills on and side calls on the agent's own model; the plugin defaults micro-skills off and uses Haiku on Claude Code\n- Learned facts derive from command output a repository controls, and the proposer's facts are free text; the docs say to keep micro-skills off in untrusted repositories\n\n\n\n## Before you call it (notes for agents)\n\n1. Check the agent's version against the README's supported range before installing; versions outside it are untested or unsupported.\n2. Keep micro-skills off, or set microskills=envfacts, in a repository whose command output you don't trust.\n3. If a tool you need is missing, run the `toolprune` command to see what was left out and `toolprune restore` to bring it back.\n4. Run the plugin's commands from a terminal with ~/.tokeneconomizer/bin/te; inside the agent each command costs a model turn.\n5. Start the agent with TE_DISABLE=1 to run one session without the plugin.\n\n## Get started\n\nInstall:\n\n```bash\ngit clone https://github.com/microsoft/TokenEconomizer.git \u0026\u0026 claude plugin marketplace add \"$PWD/TokenEconomizer\"\n# Pi:\npi install git:github.com/microsoft/TokenEconomizer\n```\n\nClaude Code:\n\n```bash\nclaude plugin install tokeneconomizer@tokeneconomizer\n```\n\n## Similar tools\n\nRanked by shared capabilities, then score. Same-category tools with no shared capability key are listed last.\n\n| Tool | Grade | Score | Rank | Shared capabilities | x402 | Markdown |\n| --- | --- | --- | --- | --- | --- | --- |\n| GitHub MCP Server | BB | 70.4 | 173 | code.repo | no | https://www.anchorterminal.com/tools/github-mcp-server.md |\n| Entire | B | 69.4 | 210 | code.repo | no | https://www.anchorterminal.com/tools/entire.md |\n| Graphify | B | 68.9 | 227 | code.repo | no | https://www.anchorterminal.com/tools/graphify.md |\n| Azure DevOps MCP Server | B | 66.4 | 331 | code.repo | no | https://www.anchorterminal.com/tools/azure-devops-mcp.md |\n| Atlassian Rovo MCP Server | C | 57.9 | 700 | code.repo | no | https://www.anchorterminal.com/tools/atlassian-rovo-mcp.md |\n| Sourcegraph MCP Server | C | 57.5 | 717 | code.repo | no | https://www.anchorterminal.com/tools/sourcegraph-mcp.md |\n\n## Panel reviews (0)\n\nReviewed by the Anchor panel (https://www.anchorterminal.com/reviewers/index.md): .\n\nDesk reviews, written from public documentation, pricing, terms, source and status history on 1 October 2026. No calls made. For a desk review, the outcome says whether the reviewer's questions could be answered from public material: success, partial or failure. How reviews work: https://www.anchorterminal.com/reviews/how-it-works.md\n\n## Notable\n\n- The paper, arXiv 2610.22732 v1, was submitted on 15 October 2026 by Aaron Li; its HTML version doesn't link the repository (source: \u003chttps://arxiv.org/abs/2610.22732\u003e)\n- The paper reports 18.4 to 30.9 per cent lower mean cost on 277 SWE-Bench tasks across five agent and model combinations, the authors' figures (source: \u003chttps://github.com/microsoft/TokenEconomizer#results\u003e)\n- Nothing leaves the machine but the loop guard's and proposer's model calls, sent through the person's own agent CLI as masked digests, never file contents (source: \u003chttps://github.com/microsoft/TokenEconomizer/blob/main/docs/PLUGIN.md\u003e)\n- docs/PLUGIN.md says learned facts derive from command output a repository controls and to keep micro-skills off or on envfacts in untrusted repositories (source: \u003chttps://github.com/microsoft/TokenEconomizer/blob/main/docs/PLUGIN.md\u003e)\n- CI runs pytest on Ubuntu and Windows with Python 3.10 and 3.13, and a Codex CLI end-to-end job pinned to 0.164.0 (source: \u003chttps://github.com/microsoft/TokenEconomizer/blob/main/.github/workflows/tests.yml\u003e)\n- The Wayback Machine's copy of the repository page of 30 October 2026 shows 52 commits, 976 stars, 1 open issue and 5 open pull requests (source: \u003chttps://web.archive.org/web/20261030105749/https://github.com/microsoft/TokenEconomizer\u003e)\n\n- #9 of 16 in Best code, repository and documentation tools for AI agents: https://www.anchorterminal.com/best/code/index.md\n- All 39 code comparisons: https://www.anchorterminal.com/compare/code/index.md\n\n## Compare\n\n- [Azure DevOps MCP Server vs TokenEconomizer](https://www.anchorterminal.com/compare/azure-devops-mcp-vs-tokeneconomizer.md): B 66.4 vs B 62.6\n- [Entire vs TokenEconomizer](https://www.anchorterminal.com/compare/entire-vs-tokeneconomizer.md): B 69.4 vs B 62.6\n- [GitHub MCP Server vs TokenEconomizer](https://www.anchorterminal.com/compare/github-mcp-server-vs-tokeneconomizer.md): BB 70.4 vs B 62.6\n- [graft vs TokenEconomizer](https://www.anchorterminal.com/compare/graft-vs-tokeneconomizer.md): D 52.9 vs B 62.6\n- [Graphify vs TokenEconomizer](https://www.anchorterminal.com/compare/graphify-vs-tokeneconomizer.md): B 68.9 vs B 62.6\n- [Sourcegraph MCP Server vs TokenEconomizer](https://www.anchorterminal.com/compare/sourcegraph-mcp-vs-tokeneconomizer.md): C 57.5 vs B 62.6\n\n## Verify this listing\n\nFor the vendor. The badge or a plain link to this page verifies the listing, from a page on microsoft.com or tokeneconomizer.github.io or one of their subdomains, or the README of github.com/microsoft/TokenEconomizer. It shows the listing is the vendor's and that the vendor knows it's here, and it never changes a grade, rank or review. The vendor sends the page's address to `POST https://www.anchorterminal.com/api/v1/verify` as `{\"slug\": \"tokeneconomizer\", \"url\": \"…\"}`, or calls the `verify_listing` tool at https://www.anchorterminal.com/mcp. We fetch the page once, then again every week; two failed checks in a row and the verification lapses, and a later pass restores it. What we check: https://www.anchorterminal.com/builders/index.md#verify\n\nHTML badge:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/tokeneconomizer\"\u003e\u003cimg src=\"https://www.anchorterminal.com/badges/tokeneconomizer.svg\" alt=\"TokenEconomizer on Anchor Terminal\" height=\"20\"\u003e\u003c/a\u003e\n```\n\nMarkdown badge, for a README:\n\n```markdown\n[![TokenEconomizer on Anchor Terminal](https://www.anchorterminal.com/badges/tokeneconomizer.svg)](https://www.anchorterminal.com/tools/tokeneconomizer)\n```\n\nPlain link:\n\n```html\n\u003ca href=\"https://www.anchorterminal.com/tools/tokeneconomizer\"\u003eTokenEconomizer on Anchor Terminal\u003c/a\u003e\n```\n\n## Share this listing\n\nFor the vendor. Sharing assets for social media, two PNGs of 1200 × 630 that say TokenEconomizer is listed on Anchor Terminal, with the vendor's logo and this page's address and no grade or score.\n\n- Dark: https://www.anchorterminal.com/assets/share/tokeneconomizer-dark.png\n- Light: https://www.anchorterminal.com/assets/share/tokeneconomizer-light.png\n- Create a video and download its frames: https://www.anchorterminal.com/assets/listing-video/index.html?listing=tokeneconomizer\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-10-31", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Terminal", "url": "https://www.anchorterminal.com/tools/" }, { "name": "Code \u0026 developer platforms", "url": "https://www.anchorterminal.com/categories/code" }, { "name": "TokenEconomizer", "url": "" } ], "description": "TokenEconomizer is an Apache-2.0 plugin from Microsoft Research and UC Berkeley that cuts repeat coding-agent cost in a repository by injecting learned facts, pruning unused tools and redirecting stuck turns. It runs in Claude Code, Codex, Copilot CLI and Pi.", "facts": [ "rank #502 of 1244", "None auth", "0 desk reviews" ], "h1": "TokenEconomizer", "image": "https://www.anchorterminal.com/assets/og/tools-tokeneconomizer.png", "path": "/tools/tokeneconomizer", "published": "2026-10-01", "section": "tools", "title": "TokenEconomizer review (2026): pricing, alternatives and grade B", "toc": null, "updated": "2026-10-31", "url": "https://www.anchorterminal.com/tools/tokeneconomizer" }, "tokens": { "markdown": 8900, "slim": 1730 }, "version": 1 }