Pi

# Fix list: Pi

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/earendil-pi, the October 2026 research run, assessed 8 October 2026. Grade B, 68.4 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Pi: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 61 out of 100, up to 6.8 more on the total

Why it scored 61: Harness reading of the framework checklist, used for the harnesses in this category. 30 for what leaves the machine by default, 20 for approvals and sandboxing, 15 for prompt-injection posture, 15 for audit and 20 for the security programme. An anonymous install and update ping to pi.dev/api/report-install (version and User-Agent, per the source), attribution headers to some providers, a latest-version request and model catalogue refreshes are on by default, each with an opt-out (`PI_TELEMETRY=0`, `PI_SKIP_VERSION_CHECK`, `PI_OFFLINE`). Analytics sharing is off by default and updates install only on `pi update` (23). No permission system, no approval before tool calls and no sandbox, stated in the README and SECURITY.md. Project trust gates a repository's `.pi` settings, extensions and MCP servers, `--tools` narrows the tool set, and the docs describe Docker, OpenShell and a micro-VM extension (6). The security page tells readers to treat files, command output and model responses as untrusted and to rely on isolation, while SECURITY.md puts prompt injection out of scope and context files such as AGENTS.md load whatever the trust decision (7). Sessions are JSONL files on disk with HTML export, MCP server logs go to `~/.pi/agent/mcp.log`, and the JSON stream records each tool call (11). SECURITY.md gives security@earendil.com and GitHub private reporting, four advisories with CVEs were published with fixes, npm releases carry provenance from trusted publishing, and a scheduled workflow runs `npm audit`. No security.txt and no bounty found (14).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Reliability, 75 out of 100, up to 5 more on the total

Why it scored 75: Read as a local package. `@earendil-works/pi-coding-agent` 1.1.0 on npm with Node.js 22.19 or newer stated, install scripts for macOS, Linux and Windows, Nix and standalone binaries (20). A public CI workflow builds, checks and tests on every push to main and runs an MCP client conformance job. The run on the commit we cloned (ce950d7, 8 October 2026) passed, but eight of the last 15 runs on main, all on 7 and 8 October, failed (15). The repository API reports 308 open issues and pull requests against 113,417 stars. Issues from new contributors are closed automatically and reviewed daily by maintainers, per CONTRIBUTING.md, so the open count understates what is reported, and we couldn't sample reply times (16). A dated changelog with Breaking Changes sections and migration steps, but the Azure provider rename shipped as a breaking change in patch release 1.0.3 on 5 October 2026, four days after 1.0.0 (9). 1.1.0, with 1.0.0 dated 1 October 2026 (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 3. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: No payment protocol in the repository or docs (0). Read with the self-hosted rule. Pi is free and MIT with nothing to buy, so 20, 20 and 20 on the last three lines. Model costs go to whichever provider the owner configures. Radius, Earendil's optional hosted gateway, is an early alpha sold by prepaid credits, and we found no public price list for it.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 4. Agent ergonomics, 76 out of 100, up to 3.9 more on the total

Why it scored 76: Harness reading of the framework line, scored on what an agent or pipeline driving it has to supply. Four default tools (read, bash, edit, write) and a short system prompt. MCP tools default to `codemode` exposure, which keeps them out of the model's tool declarations, and `tool_search` loads deferred ones on demand (23). Automatic compaction, agent-level retry with three attempts and exponential backoff, and truncated tool output saved to a file. We found no turn or step limit in the CLI or settings reference (13). `--mode json` streams JSONL events ending in `agent_settled` and `--mode rpc` returns `success: false` with an error string. Print mode exits nonzero on a failed response, JSON mode doesn't (16). Sessions are stored as trees and continue, fork or take a fixed ID with `--continue`, `--fork` and `--session-id`. File changes have no built-in undo, and a git checkpoint is an example extension (14). It needs a provider key or subscription login before the first run, and the SDK is TypeScript only, with RPC for other languages (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Transparency & trust, 64 out of 100, up to 3.2 more on the total

Made of editorial 64, provenance 63.

Why it scored 64: MIT (30). No privacy policy for pi.dev or the CLI was found (pi.dev/privacy and earendil.com/privacy return 404). The docs say where sessions and credentials are stored and what `/share` and `/bug` upload. The Radius alpha terms of 1 September 2026 name Earendil Works Co. and refer to a Privacy Notice we couldn't locate (10). Breaking changes are listed per release with migration steps, with no deprecation policy or notice period found (8). Install telemetry and the version check are documented in the settings and environment-variable pages with opt-outs, though the docs don't list the fields sent (16).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: pi.dev, registered 2024-01-30 (2 years) (7 of 15)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 84 out of 100, up to 2.6 more on the total

Why it scored 84: Harness reading, as for the other harnesses. JSON Schemas for settings, models, keybindings and themes ship in the package, and the JSON event stream and every RPC command are documented record by record, but as Markdown and TypeScript types with no machine-readable spec for the RPC protocol (18). pi.dev/llms.txt returns 404. Each docs page answers `Accept: text/markdown` with Markdown, and the docs ship inside the npm package (8). The integration page has a table saying when to use interactive, print, JSON, RPC or the SDK, and the MCP page gives a typical use for each exposure setting (17). Enumerated modes, thinking levels and exposure values, typed settings with defaults (13). A Python RPC client, a typed TypeScript client and about 75 example extensions, with command errors and parse errors shown as records (13). A dated changelog in the repository and at pi.dev/changelog with an RSS feed (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Maintenance & community, 87 out of 100, up to 1.1 more on the total

Why it scored 87: v1.1.0 on 7 October 2026, one day before this check (30). 33 tagged releases since 10 July 2026 (20). Commits land daily and the changelog credits issues and outside pull requests in most releases. New contributors' issues are closed automatically, reviewed daily, and get no reply if they miss the quality bar, per CONTRIBUTING.md. Reply times weren't sampled because the GitHub API rate limit was reached (16). npm, the pi.dev installer, Nix and standalone binaries track each release, and npm publishes through GitHub Actions trusted publishing (14). Direct dependencies are pinned, a scheduled workflow runs `npm audit`, and CI passed on the commit we cloned after eight failures in the previous 14 runs (7).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 2026-06-08. GHSA-jfgx-wxx8-mp94 (CVE-2026-54328, high). Temporary extension installs used predictable paths under the system temp directory, so another local user on a shared Linux host could plant extension code that Pi then loaded. Fixed in 0.78.1. Fixed, published and four months old, -2. https://github.com/earendil-works/pi/security/advisories/GHSA-jfgx-wxx8-mp94
- 2026-06-08. GHSA-mqxh-6gq7-558m (CVE-2026-54325, medium). Before 0.79.0 Pi loaded a repository's `.pi` settings and extensions without asking, so starting it in a cloned repository could run that repository's code. Fixed in 0.79.0 with project trust. Fixed and published, -2. https://github.com/earendil-works/pi/security/advisories/GHSA-mqxh-6gq7-558m
- 2026-06-08. GHSA-7v5m-pr3q-6453 (CVE-2026-54326, low, XSS in HTML session exports) and GHSA-r95r-rj6r-c39x (CVE-2026-54327, low, a race in `auth.json` permissions). Both fixed in 0.78.1. Recorded without a deduction. https://github.com/earendil-works/pi/security/advisories

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: issue and pull request reply times, and separate open issue and pull request counts. The GitHub API rate limit was reached, so only the combined count of 308 was read
- unchecked: why eight of the last 15 CI runs on main failed. The run logs weren't read
- The Privacy Notice the Radius alpha terms refer to wasn't found, and no privacy policy for pi.dev or the CLI was found
- The legal entity is unclear. The pi.dev footer reads Earendil Inc. and the Radius terms name Earendil Works Co.
- No public price list for Radius credits was found. Radius is optional and wasn't graded
- The date of Pi's first release wasn't established. The changelog starts at 0.10.0 on 25 November 2025 and the repository was created on 9 August 2025
- The scout said MCP wasn't mentioned in the README. Pi has had a built-in MCP client since 0.99.0 on 29 September 2026, documented in docs/mcp.md

## Weaknesses

- No permission system or sandbox. Tool calls run with the user's rights and without an approval prompt
- An anonymous install and update ping to pi.dev and a latest-version request are on by default
- Four advisories published on 8 June 2026, one rated high and one medium, all fixed by 0.79.0
- No privacy policy for pi.dev or the CLI found, and pi.dev has no security.txt
- Eight of the last 15 CI runs on main failed on 7 and 8 October 2026, and a breaking provider rename shipped in patch release 1.0.3

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Run Pi inside a container or VM for unattended work. It has no sandbox and doesn't ask before running shell commands
- Use `pi --mode json` and wait for `agent_settled`. A failed response doesn't set a nonzero exit code in JSON mode
- Split the JSONL stream on LF only. Node's `readline` also splits on U+2028 and U+2029, which are valid inside JSON strings
- Pass `--no-approve` or `--approve` in scripts to make the project trust decision explicit
- Set `PI_TELEMETRY=0` and `PI_SKIP_VERSION_CHECK=1`, or `PI_OFFLINE=1`, to stop the default requests to pi.dev

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.