getappmap/AppMap Behavioral Review

Maintain a project's AppMap gold-trace baseline and review the runtime-behavior change between two revisions. The work is performed by an AI coding agent (Claude Code, headless) running the getappmap/skills `appmap-gold-traces` and `appmap-review` skills. Updated gold traces (and any labels/tests the agent adds) are committed and pushed to the PR branch; the review is posted as a PR comment and job summary.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
agentWhich agent runtime executes the skills: `claude` (Claude Code, default) or `copilot` (GitHub Copilot CLI). claude
anthropic-api-keyAPI key used to authenticate the Claude Code agent. Required when `agent` is `claude`; ignored when `agent` is `copilot`. no
copilot-tokenGitHub token with Copilot access, used when `agent` is `copilot`. The default workflow `GITHUB_TOKEN` cannot use Copilot, so supply a PAT (or Copilot-enabled token) via a secret. Falls back to `github-token` if unset. no""
github-tokenToken used to push gold-trace updates to the PR branch and to post the review comment. Needs `contents: write` and `pull-requests: write`. Defaults to the workflow token. ${{ github.token }}
base-revisionThe baseline revision the review compares against. Defaults to the PR base ref (`GITHUB_BASE_REF`). A revision is any git ref (SHA, branch, tag). ${{ github.base_ref }}
head-revisionThe head revision to review. Defaults to the commit that triggered the run (`GITHUB_SHA`). Note: the action first records + commits gold-trace updates on the head branch, so the review reads the freshly-blessed head traces. ${{ github.sha }}
gold-traces-dirManaged gold-traces directory, relative to `working-directory` (the engine's `--dir`).gold_traces
working-directoryDirectory the agent and gold-traces engine run from. Set to a package root in a monorepo. Defaults to the repository root. .
comment-tagDistinguishes this run's sticky PR comment when the action runs more than once per PR (e.g. a matrix). Each distinct tag gets its own comment, updated in place on re-runs; set it to a stable per-entry key such as the matrix entry name. Defaults to `working-directory`, so monorepo matrix runs get separate comments automatically. `.` (the `working-directory` default) means untagged. ""
skills-repoGit URL of the skills repository to install.https://github.com/getappmap/skills.git
skills-refBranch, tag, or SHA of the skills repository to pin.main
modelPrimary model for the selected agent (passed as `--model` to both `claude` and `copilot`). Empty uses the agent's default. Set it to run cheaper, e.g. `claude-sonnet-4-5` instead of Opus 4.8. ""
mini-modelThe small/fast ("mini") model used for lightweight background work. Supported by the `claude` agent (via `ANTHROPIC_SMALL_FAST_MODEL`); the `copilot` agent has no such concept and will warn and ignore it. Empty uses the agent's default. ""
appmap-cli-versionVersion of the AppMap CLI to install from getappmap/appmap-js GitHub releases (e.g. `3.196.2`). Takes precedence over an `appmap` binary already on PATH. Empty uses the PATH binary if present, else installs the latest release. ""
node-versionNode.js version the action needs (the agent CLIs and the bundled usage script run on Node). If the workflow already provides `node` at this major version or newer it is used as-is; otherwise the action installs this version. Override if the default causes trouble on your setup. 22
commit-messageCommit subject used when pushing gold-trace (and related) updates. `[skip ci]` is appended automatically.chore(gold-traces): update behavioral baseline
namedescription
report-filePath to the generated Markdown review report.
gold-traces-updatedtrue if the agent produced committable changes that were pushed.
modelsComma-separated model ids the agent reported using.
cost-usdTotal cost in USD, as reported by Claude Code. Empty for the copilot agent.
premium-requestsTotal premium requests, as reported by the Copilot CLI. Empty for the claude agent.
input-tokensTotal input tokens (including cache reads/writes) reported by the agent. Empty when not reported.
output-tokensTotal output tokens reported by the agent. Empty when not reported.
duration-msTotal agent wall time in milliseconds, summed across the update and review runs.