More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Three Axes Framework - Supporter Edition

    1

    Keep AI-assisted code reviews tied to the change you inspected: decisions, human answers and check evidence saved locally, with source comparisons before reuse.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Three Axes Framework - Supporter Edition

    Three Axes Framework - Supporter Edition

    Example session with this skill installed

    Use the Windows onboarding fix.

    Agreed scope: public Three Axes PR #8, cd0bc90 → b73c4c4, hooks/lib/setup.mjs. Record the review locally, run the targeted onboarding checks, and revisit the saved evidence. Settings: high mastery, medium consequence, balanced intent, declared from this conversation.

    • Read your context and instructions
    • Compiled the three axes framework

    Recorded locally with the released Supporter 1.7.3 runtime. This is a real review of public PR #8; live Agensi delivery/licensing has not been exercised.

    Selected source: luxsolari/three-axes-framework, cd0bc9066ad064c58a6981d93a1c4d34107a018b → b73c4c495ac87454c6544a11f34f152f151ea6b5. File: hooks/lib/setup.mjs. Working-tree/repository-context capture enabled.

    Axes: mastery high; consequence medium; intent balanced. These were explicitly selected for this demo and recorded as conversation declarations.

    Decision: Generate a PowerShell command on Windows and a POSIX command elsewhere; invoke the current Node executable through its quoted absolute path. The generated command is explicitly tied to the shell named in setup guidance.

    Control flow: setupContext offers the global/project writer command; setupCommand quotes executable, writer and target. isSetupCommand extracts the three allowed axis values and permits only an exact regenerated command for one of the persistent profile paths.

    Invariant: Executable, writer and profile paths must arrive as literal arguments, including apostrophes, smart single quotes, dollar expressions and separators. The user may replace only validated axis values; extra commands or another target do not match the allowlist.

    Failure path: POSIX apostrophe escaping is incompatible with native PowerShell. A path may split or be interpreted as shell syntax; selecting the wrong shell can likewise prevent the save. Missing or invalid profile state keeps project work gated.

    Evidence scope: This record reviews the real released public change cd0bc90 to b73c4c4, limited to hooks/lib/setup.mjs, in an isolated clean checkout. Targeted actual-shell tests will be observed with matching before/after source bindings; no previous CI output is copied in as a new observation.

    Review limits: The selected-file review does not establish behavior of every caller or host UI. This local native runtime run does not establish live Agensi delivery/licensing or certify the developer's understanding. Saved evidence must be compared again before reuse; comparison alone does not rerun tests.

    Implementation question
    Context: we are reviewing the free plugin's first-run profile writer as source code; Supporter is saving our review, not installing hooks. On Windows, setup may live in a folder such as O'Brien's tools, and the target path can contain dollar signs or other shell characters. The fix quotes the executable, writer and profile path for PowerShell and still accepts only an exactly regenerated setup command. In your own words, what should remain true about those three arguments when the command executes, and why would accepting any command that merely mentions configure-profile.mjs be too permissive?

    Illustrative answer — assistant-authored for this demo:

    The Node executable, writer script and profile path must reach the command as the exact intended arguments. Spaces, apostrophes, dollar signs and other shell characters belong to those paths: they must not split an argument, expand a variable or introduce another command. Windows needs PowerShell literal quoting and the call operator for the quoted executable; POSIX shells use their own quoting rules.

    A filename match only proves that configure-profile.mjs appears somewhere in the text. It could appear in a different invocation, target another file, use invalid axis values or be chained with another command. Regenerating the exact allowed invocation from the approved profile path and validated axis values keeps the onboarding exception limited to the intended writer call. It is a scoped workflow guard, not a general shell sandbox.

    This sample is saved as an assistant interpretation. The genuine-human-answer slot remains open; the sample is not attributed to Lux and does not establish human comprehension.

    Actual host check

    ✔ setup writer is allowed only for exact commands and approved paths (2053.2204ms)
    ✔ global onboarding creates the selected baseline and reports its values (722.3758ms)
    ✔ onboarding preserves quoted installation and profile paths in the declared shell (817.3057ms)
    ✔ PowerShell 7 executes the same literal onboarding command (549.4363ms)
    ℹ tests 4
    ℹ suites 0
    ℹ pass 4
    ℹ fail 0
    ℹ cancelled 0
    ℹ skipped 0
    ℹ todo 0
    ℹ duration_ms 4210.1394
    

    The check ran in the selected checkout. Source/context bindings matched before and after; the invocation, exit status, output and observation time were saved. These fixtures exercise the source project's Windows PowerShell and PowerShell 7 paths; Node/PowerShell are requirements of this example check, not of every Supporter review.

    Revisit: a separate runtime process reopened the saved review and returned bindings_match. It did not rerun the historical tests.

    Controlled stale-source demonstration: I temporarily appended one labelled comment to the selected file in the isolated checkout. A revisit returned source_changed (exit 3), and an attempted map mutation was refused with committed=false. The saved record bytes stayed unchanged. I restored exactly the bytes written by this probe; a fresh comparison returned bindings_match. This edit was a deliberate demo probe, not a naturally occurring regression or a change to the released plugin.

    Result: one durable local review and readable report, six map categories, one implementation question with a clearly labelled assistant-authored sample answer, an actual check observation, and a tested stale-source refusal. The source was restored. No comprehension score or merge authorization was produced.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    The code can pass its tests while the person responsible for it still cannot explain the change.

    Three Axes Framework calibrates AI assistance to your familiarity with the work, the consequences of a mistake, and whether you are optimizing for delivery or learning. Supporter adds a local review record tied to the Git state you inspected.

    Keep the reasoning with the change

    Select the repository, base/head revisions and relevant files. Build a map of decisions, control flow, invariants, failure paths, evidence and uncertainties. Ask questions in the context of that code and keep actual human answers separate from assistant interpretations.

    Your agent runs the checks. Supporter records their declared invocation, output and before/after source bindings, keeping host-observed evidence distinct from copied claims. Save a local JSON record and readable report, then reopen them in another session.

    Revisit before relying on old evidence

    The runtime compares the selected source and captured repository context. Changed or uncheckable bindings preserve the history and prevent new entries from being silently attached to another state. Matching bindings do not rerun historical checks or establish that the code is bug-free.

    The free framework stays free

    The complete open Three Axes Framework remains free and MIT, including its ownership guidance, profiles, presets and native integrations. Supporter adds the saved records and revisit workflow. Buying this edition also supports continued development.

    Requirements and boundaries

    Requires Python 3.11+, a complete nonsparse local Git checkout, process/filesystem tools and durable user-controlled record storage. Project checks may need additional tools. Buyers do not need private source-repository access.

    Agensi delivers a text skill and Python runtime. Run shipped scripts unchanged in temporary storage and retain records separately. This edition supplies no native hooks, automatic session injection, registered slash commands, cloud storage or background monitoring.

    Entries are explicit; there is no automatic chat capture. Source comparisons validate supplied consistency, not identity or authenticity. The workflow gives no comprehension score, authorship attestation or merge authorization.

    Source capture refuses shallow/partial clones, sparse checkouts, symlinks/junctions and unsupported index states. Bindings are checked at observation points, not continuously or atomically; ignored files and external environment are excluded. Live Agensi delivery/licensing remains unverified until the approved listing is exercised.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 21 days ago

    Requires

    Python 3.11+
    Git
    Terminal access
    Durable storage for review records
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Environment Variables

    The local runtime reads the selected repository and captured tracked/non-ignored working-tree context, and saves explicit review records in a separately selected durable location. Environment variables are used for local Git/Python isolation, not sent over the network. The host executes any separately authorized checks. No automatic chat capture or cloud persistence.

    Listed21 days ago
    Updated12 days ago

    What's inside

    Frequently Asked Questions

    This skill eliminates 'vibes-based' coding by enforcing a structured review process that links technical decisions to human-verified evidence and Git revisions, preventing architectural drift and undocumented technical debt.

    The skill requires Python 3.11+ and Git; it is designed to work with any AI agent that has filesystem access and the ability to execute Python scripts in an isolated environment.

    Yes, the framework includes an Ownership Mapping feature that allows you to document intent and mastery levels for specific code paths, making it ideal for capturing knowledge during complex refactoring or legacy onboarding.

    You will receive the core Three Axes logic, the `run_review.py` execution launcher, and the necessary schemas for recording revision-linked evidence and technical invariants.

    The framework relies on a user-defined persistent storage location to save its records, ensuring that your architectural evidence and integrity logs remain available across different sessions and revisions.

    New in Code Quality & Review