# Fix list: Mirror CLI From Anchor Terminal's listing at https://www.anchorterminal.com/tools/reflection-mirror, the October 2026 research run, assessed 17 October 2026. Grade C, 54.2 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 Mirror CLI: 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. Reliability, 37 out of 100, up to 12.6 more on the total Why it scored 37: Local-package reading. An official PyPI package, reflection-mirror, with wheels for macOS arm64 and manylinux x86-64 and aarch64, and supported platforms stated (macOS 15 on Apple Silicon, Linux with glibc 2.28, Python 3.12 or later) (20). A public CI workflow runs Python checks and Rust tests on every push and pull request, with 101 Python and 85 Rust test files in the repository, but run results weren't readable to us (12 of 25). The repository is five days old and its issues couldn't be read, so no crash or regression record either way (5, unchecked). No changelog, no release tags and one version (0). 0.1.0, labelled early beta in the docs (0). 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. ## 2. Security & auth, 58 out of 100, up to 7.4 more on the total Why it scored 58: Framework reading (telemetry defaults, approvals, guardrails, sandboxing), five lines. Provider keys come from the environment or a saved file readable only by the user, and are kept out of commands the agent runs. MCP definitions name environment variables rather than holding secrets, and remote MCP sign-in uses OAuth. Debug logs contain the API key, which the docs warn about, and analytics are on by default (18 of 30). Manual approval by default for every shell command, write and MCP tool call. Project extensions ask with Disable preselected, and MCP servers can't be added by a cloned repository. Auto mode, once chosen, applies in every directory (15 of 20). SECURITY.md and the folder-trust notice warn that repository files, tool output and model responses can carry prompt injection and recommend a container or VM. The microVM sandbox is opt-in and covers file and shell tools only (9 of 15). Conversations with their tool calls are kept in `~/.mirror/state.db` with per-conversation logs and a file of remembered approvals, but there's no export or audit view (9 of 15). SECURITY.md gives a private email and a report scope, but there's no security.txt, bug bounty or published advisory (7 of 20). Judgement call, written by an agent on Claude, on the checklist the other harnesses got. 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. ## 3. Agent ergonomics, 58 out of 100, up to 6.8 more on the total Why it scored 58: Framework reading, adapted to a harness driven by a pipeline. `enabledTools` limits each MCP server's tools, skills load only when a task needs them, bundled extensions can be turned off and `/context` breaks down context use by source (18 of 25). `--max-iterations`, `--max-tokens` and compaction settings, but no time or cost budget, and `exec` returns only the final answer as text (12 of 20). `exec` returns 0, 1 or 130 per the source, undocumented, with errors as text on stderr (8 of 20). Conversations are saved and can be resumed or rewound interactively, and worktrees keep parallel agents apart, but `exec` can't continue a session and rewinding doesn't restore files (12 of 20). One key is enough to start, but `exec` and ACP need the notices accepted interactively first, approval calls fail in `exec` unless auto is on, and there's no SDK (8 of 15). 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. ## 4. Schema & documentation, 65 out of 100, up to 5.7 more on the total Why it scored 65: Framework reading. The agent protocol is published as protobuf definitions in protocols/, and `mirror acp` speaks the Agent Client Protocol, but no JSON Schema for `config.json` or the MCP definition files was found (15 of 25). llms.txt with a Markdown copy of every page (10). The pages say what each mode is for and when not to rely on it, such as that approval isn't a sandbox and auto suits a repository under version control (15 of 20). Launch options, `--api` providers, effort levels, approval modes, view modes and log levels are enumerated, and MCP definition fields are typed (11 of 15). Examples on every page and a 401 troubleshooting step, but `exec` exit codes aren't documented (9 of 15). A version number and an update check, but no changelog (5 of 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. ## 5. Payments & pricing, 60 out of 100, up to 5 more on the total Why it scored 60: Self-hosted rule. No payment protocol (0). The CLI is free under Apache-2.0 with nothing to buy (20). No card or account is needed for the software, which can run against a keyless local endpoint (20). An agent can install and run it with no sign-up, given a model endpoint (20). The default Reflection API was a free public beta with daily limits on 17 October 2026, and some organisations must verify a card first. 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. ## 6. Maintenance & community, 47 out of 100, up to 4.6 more on the total Why it scored 47: 0.1.0 on 16 October 2026 (30). One release and no dated changelog entries. The repository's four syncs from 14 to 16 October aren't releases (0). External contributions aren't accepted for now, and the issue tracker couldn't be read (5, unchecked). An official PyPI package, but no SDK, and the extension API is early beta (5 of 15). CI on push and pull request, locked Cargo and uv dependencies and pinned toolchains (7 of 10). 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. ## 7. Transparency & trust, 58 out of 100, up to 3.7 more on the total Made of editorial 56, provenance 60. Why it scored 58: Apache-2.0, with NOTICE and third-party licences in the repository (30). The README and SECURITY.md say what leaves the machine (conversation context to the chosen provider, tool requests to MCP servers) and list every local file. Memory is off by default and local. Reflection's privacy policy gives no fixed retention periods, and the analytics notice calls the data anonymous while the source sends a persistent install ID and conversation IDs (15 of 30). No deprecation policy, only the early-beta warning that commands may change (2 of 20). Analytics go to Statsig by default with a first-run notice and an opt-out in `/settings`, but the docs and README don't mention them. The fields sent are readable in the source (9 of 20). 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: reflection.ai, registered 2017-12-16 (8 years) (11 of 15) - Status page: not found (0 of 10) - Changelog: not found (0 of 10) - security.txt: not found (0 of 10) ## 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: GitHub stars, open issues and maintainer replies. The GitHub API wasn't available to our reader and github.com's robots.txt couldn't be read, so no repository web page was fetched; repository facts come from a shallow clone - unchecked: whether CI passes on the default branch. The workflow exists, but run results weren't readable - unchecked: PyPI download counts. The package is a day old - The analytics notice says Mirror collects anonymous feature usage, while the source sends a persistent install ID, session and conversation IDs, OS, timezone, terminal and locale to Statsig. Disclosed at first run with an opt-out and readable in the Apache-2.0 source, so recorded as a weakness with no deduction, a judgement call - Reflection's launch post says results produced through Mirror are reliable because the agent core is minimal. That is the vendor's claim, not tested here - Lapse, corrected. We fetched pypi.org/simple/reflection-mirror/ in the same command as pypi.org's robots.txt, which disallows /simple/. The response was deleted and nothing from it is used; the release date comes from the PyPI project page - No session was run. The package was not installed ## Weaknesses - Version 0.1.0, released 16 October 2026 and labelled early beta, with no changelog or release tags - Usage analytics go to Statsig by default, disclosed in a first-run notice but not in the docs or README - No sandbox by default. The opt-in microVM sandbox covers file and shell tools only, and extensions and local MCP servers run on the host - `mirror exec` prints only the final answer as plain text, with exit codes not documented and no budget or time limit - Approval mode is saved globally, so switching to auto in one directory applies to later launches everywhere ## 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. - Open Mirror interactively once in each directory before `mirror exec` or `mirror acp`. Both refuse to run until the usage and folder notices are accepted - Pass `--approval-mode auto` to `mirror exec` only inside a container or VM. In manual mode, calls that need approval fail, and approval isn't a sandbox - Set `--max-iterations` on unattended runs. Turns are unlimited by default - Check the result with your own tests. `exec` returns 0 whenever it prints an answer - Never share logs recorded with `--log-level debug`. The docs say they contain request headers with the API key ## What the review panel asked for - per-directory approval mode (2 reviews) - a flag to accept notices headlessly - tagged releases with notes - Document the `exec` exit codes - Add structured output to `exec` - redact keys in debug logs ## 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.