action.yml

name: 'CodeBoarding Action'
description: 'Review pull request architecture changes or sync a versioned CodeBoarding baseline.'
author: 'CodeBoarding'
branding:
  icon: 'git-pull-request'
  color: 'blue'
inputs:
  mode:
    description: 'review posts a PR architecture diff; sync updates the versioned analysis baseline.'
    required: false
    default: 'review'
  llm:
    description: 'Required. Where analysis credentials come from: hosted, license, or a provider name (anthropic, aws_bedrock, cerebras, deepseek, glm, google, kimi, litellm, ollama, openai, openrouter, opper, orcarouter, requesty, vercel).'
    required: true
  license_key:
    description: 'CodeBoarding license key. Required by llm: license; optional alongside a provider key.'
    required: false
    default: ''
  # Anthropic - selected by llm: anthropic
  anthropic_api_key:
    description: 'Anthropic API key, used when llm is anthropic. Sets ANTHROPIC_API_KEY.'
    required: false
    default: ''
  # AWS Bedrock - selected by llm: aws_bedrock
  aws_bedrock_api_key:
    description: 'AWS Bedrock API key, used when llm is aws_bedrock. Sets AWS_BEARER_TOKEN_BEDROCK.'
    required: false
    default: ''
  aws_bedrock_region:
    description: 'AWS Bedrock region, used when llm is aws_bedrock. Sets AWS_DEFAULT_REGION.'
    required: false
    default: ''
  # Cerebras - selected by llm: cerebras
  cerebras_api_key:
    description: 'Cerebras API key, used when llm is cerebras. Sets CEREBRAS_API_KEY.'
    required: false
    default: ''
  # DeepSeek - selected by llm: deepseek
  deepseek_api_key:
    description: 'DeepSeek API key, used when llm is deepseek. Sets DEEPSEEK_API_KEY.'
    required: false
    default: ''
  deepseek_base_url:
    description: 'DeepSeek endpoint, used when llm is deepseek. Sets DEEPSEEK_BASE_URL.'
    required: false
    default: ''
  # GLM - selected by llm: glm
  glm_api_key:
    description: 'GLM API key, used when llm is glm. Sets GLM_API_KEY.'
    required: false
    default: ''
  glm_base_url:
    description: 'GLM endpoint, used when llm is glm. Sets GLM_BASE_URL.'
    required: false
    default: ''
  # Google Gemini - selected by llm: google
  google_api_key:
    description: 'Google Gemini API key, used when llm is google. Sets GOOGLE_API_KEY.'
    required: false
    default: ''
  # Kimi - selected by llm: kimi
  kimi_api_key:
    description: 'Kimi API key, used when llm is kimi. Sets KIMI_API_KEY.'
    required: false
    default: ''
  kimi_base_url:
    description: 'Kimi endpoint, used when llm is kimi. Sets KIMI_BASE_URL.'
    required: false
    default: ''
  # LiteLLM - selected by llm: litellm
  litellm_api_key:
    description: 'LiteLLM API key, used when llm is litellm. Sets LITELLM_API_KEY.'
    required: false
    default: ''
  litellm_base_url:
    description: 'LiteLLM endpoint, used when llm is litellm. Sets LITELLM_BASE_URL.'
    required: false
    default: ''
  # Ollama - selected by llm: ollama
  ollama_api_key:
    description: 'Ollama API key, used when llm is ollama. Sets OLLAMA_API_KEY.'
    required: false
    default: ''
  ollama_base_url:
    description: 'Ollama endpoint, used when llm is ollama. Sets OLLAMA_BASE_URL.'
    required: false
    default: ''
  # OpenAI - selected by llm: openai
  openai_api_key:
    description: 'OpenAI API key, used when llm is openai. Sets OPENAI_API_KEY.'
    required: false
    default: ''
  openai_base_url:
    description: 'OpenAI endpoint, used when llm is openai. Sets OPENAI_BASE_URL.'
    required: false
    default: ''
  # OpenRouter - selected by llm: openrouter
  openrouter_api_key:
    description: 'OpenRouter API key, used when llm is openrouter. Sets OPENROUTER_API_KEY.'
    required: false
    default: ''
  # Opper - selected by llm: opper
  opper_api_key:
    description: 'Opper API key, used when llm is opper. Sets OPPER_API_KEY.'
    required: false
    default: ''
  # OrcaRouter - selected by llm: orcarouter
  orcarouter_api_key:
    description: 'OrcaRouter API key, used when llm is orcarouter. Sets ORCAROUTER_API_KEY.'
    required: false
    default: ''
  # Requesty - selected by llm: requesty
  requesty_api_key:
    description: 'Requesty API key, used when llm is requesty. Sets REQUESTY_API_KEY.'
    required: false
    default: ''
  requesty_base_url:
    description: 'Optional Requesty endpoint, used when llm is requesty. Sets REQUESTY_BASE_URL.'
    required: false
    default: ''
  # Vercel AI Gateway - selected by llm: vercel
  vercel_api_key:
    description: 'Vercel AI Gateway API key, used when llm is vercel. Sets VERCEL_API_KEY.'
    required: false
    default: ''
  vercel_base_url:
    description: 'Vercel AI Gateway endpoint, used when llm is vercel. Sets VERCEL_BASE_URL.'
    required: false
    default: ''
  model:
    description: 'Optional model used for both analysis and parsing.'
    required: false
    default: ''
  agent_model:
    description: 'Optional analysis-model override. Takes precedence over model.'
    required: false
    default: ''
  parsing_model:
    description: 'Optional parsing-model override. Takes precedence over model.'
    required: false
    default: ''
  depth_cap:
    description: 'Positive integer analysis depth cap, used for both review and sync, including full fallbacks.'
    required: false
    default: '2'
  github_token:
    description: 'Token used for comments and sync delivery.'
    required: false
    default: ${{ github.token }}
  codeboarding_analysis_location:
    description: >-
      Where the analysis of the synced branch (the one branch the sync workflow's
      push trigger lists) is stored. Sync saves it there and review reads its base
      from there, so set the same value for both.
      codeboarding_branch (default): on codeboarding/baseline, an orphan branch of its
      own with one commit per sync; the synced branch is never written. Sync then runs
      on the repository's default branch only.
      in_place: committed as .codeboarding/ on the synced branch itself.
    required: false
    default: 'codeboarding_branch'
  warmstart_retention_days:
    description: 'Days to keep the reusable analysis. Only the next run reads it.'
    required: false
    default: '1'
  # Deprecated inputs. scripts/action/old_api_migrator.sh is the only reader.
  target_branch:
    description: 'Deprecated: sync analyzes the branch the run is on. Accepted only when it names that branch.'
    deprecationMessage: 'target_branch is deprecated; sync analyzes the branch the run is on.'
    required: false
    default: ''
  sync_strategy:
    description: 'Deprecated: use codeboarding_analysis_location (push is in_place, pull_request is codeboarding_branch).'
    deprecationMessage: 'sync_strategy is deprecated; use codeboarding_analysis_location.'
    required: false
    default: ''
  force_full:
    description: 'Deprecated and ignored: sync always continues from the nearest saved analysis.'
    deprecationMessage: 'force_full is deprecated and ignored.'
    required: false
    default: ''
outputs:
  llm_tier:
    description: 'Resolved credential tier: hosted, license, byok, or byok+license.'
    value: ${{ steps.llm.outputs.tier }}
  llm_provider:
    description: 'Provider this run used for analysis.'
    value: ${{ steps.llm.outputs.provider }}
  llm_config_error:
    description: 'Stable code for a credential configuration failure, empty when configured.'
    value: ${{ steps.llm.outputs.error }}
  diagram_md:
    description: 'Path to the rendered Mermaid review diagram.'
    value: ${{ steps.review_render.outputs.diagram_md }}
  n_changed:
    description: 'Number of changed components in review mode.'
    value: ${{ steps.review_render.outputs.n_changed }}
  truncated:
    description: 'Whether the review graph was reduced to fit GitHub limits.'
    value: ${{ steps.review_render.outputs.truncated }}
  review_artifact_url:
    description: 'URL of the uploaded review analysis artifact.'
    value: ${{ steps.upload_review_artifact_dotcom.outputs.artifact-url }}
  seed_source:
    description: 'Which state the review head analysis grew from: pr-chain or base.'
    value: ${{ steps.review_analyze.outputs.seed_source }}
  merge_base_sha:
    description: 'Merge base used as the review comparison baseline.'
    value: ${{ steps.guard.outputs.merge_base_sha }}
  analysis_mode:
    description: 'Whether sync used incremental or full analysis.'
    value: ${{ steps.sync_analyze.outputs.analysis_mode }}
  files_written:
    description: 'Number of persisted analysis artifacts produced by sync.'
    value: ${{ steps.sync_commit.outputs.files_written }}
  committed:
    description: 'Whether sync delivered a baseline commit.'
    value: ${{ steps.sync_commit.outputs.committed }}

runs:
  using: 'composite'
  steps:
    - name: Translate deprecated inputs
      id: inputs
      shell: bash
      env:
        MODE: ${{ inputs.mode }}
        REF_NAME: ${{ github.ref_name }}
        ANALYSIS_LOCATION: ${{ inputs.codeboarding_analysis_location }}
        OLD_TARGET_BRANCH: ${{ inputs.target_branch }}
        OLD_SYNC_STRATEGY: ${{ inputs.sync_strategy }}
        OLD_FORCE_FULL: ${{ inputs.force_full }}
      run: "$GITHUB_ACTION_PATH/scripts/action/old_api_migrator.sh"

    - name: Resolve event
      id: guard
      shell: bash
      env:
        GH_TOKEN: ${{ inputs.github_token }}
        GH_ENTERPRISE_TOKEN: ${{ inputs.github_token }}
        GH_HOST: ${{ github.server_url }}
        MODE: ${{ inputs.mode }}
        EVENT: ${{ github.event_name }}
        REF_NAME: ${{ github.ref_name }}
        REF_TYPE: ${{ github.ref_type }}
        HEAD_AUTHOR_EMAIL: ${{ github.event.head_commit.author.email }}
        ANALYSIS_LOCATION: ${{ steps.inputs.outputs.analysis_location }}
        COMMENT_BODY: ${{ github.event.comment.body }}
        AUTHOR_ASSOCIATION: ${{ github.event.comment.author_association }}
        ISSUE_PR_URL: ${{ github.event.issue.pull_request.url }}
        EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
        PULL_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
        PULL_BASE_SHA: ${{ github.event.pull_request.base.sha }}
        PULL_BASE_REF: ${{ github.event.pull_request.base.ref }}
        PULL_HEAD_REPO: ${{ github.event.pull_request.head.repo.full_name }}
        PULL_BASE_REPO: ${{ github.event.pull_request.base.repo.full_name }}
        REPOSITORY: ${{ github.repository }}
        DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
      run: "$GITHUB_ACTION_PATH/scripts/action/resolve-github-event.sh"

    - name: Acknowledge review command
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.event == 'issue_comment'
      continue-on-error: true
      shell: bash
      env:
        GH_TOKEN: ${{ inputs.github_token }}
        GH_ENTERPRISE_TOKEN: ${{ inputs.github_token }}
        GITHUB_SERVER_URL: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
        COMMENT_ID: ${{ github.event.comment.id }}
      run: GH_HOST="${GITHUB_SERVER_URL#*://}" gh api -X POST "repos/${REPOSITORY}/issues/comments/${COMMENT_ID}/reactions" -f content=eyes >/dev/null

    # Ahead of the credential check rather than after the checkout, because that check
    # is a Python program. Hosted runners ship a system python3 and would not notice;
    # a self-hosted runner without one would fail a perfectly valid configuration. It
    # costs a couple of seconds and still leaves the check ahead of the engine install,
    # which is the part worth failing before.
    - name: Setup Python
      if: steps.guard.outputs.skip != 'true'
      uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
      with:
        python-version: '3.12'

    # Resolved before the checkout and the engine install, so a workflow that names a
    # provider it has no key for fails within seconds, naming the input to fix, instead
    # of running to completion on credentials it never asked for. Nothing downstream
    # re-decides this: the plan it writes is what the analysis runs with.
    - name: Check LLM configuration
      id: llm
      if: steps.guard.outputs.skip != 'true'
      continue-on-error: true
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CB_IN_LLM: ${{ inputs.llm }}
        CB_IN_LICENSE_KEY: ${{ inputs.license_key }}
        CB_IN_ANTHROPIC_API_KEY: ${{ inputs.anthropic_api_key }}
        CB_IN_AWS_BEDROCK_API_KEY: ${{ inputs.aws_bedrock_api_key }}
        CB_IN_AWS_BEDROCK_REGION: ${{ inputs.aws_bedrock_region }}
        CB_IN_CEREBRAS_API_KEY: ${{ inputs.cerebras_api_key }}
        CB_IN_DEEPSEEK_API_KEY: ${{ inputs.deepseek_api_key }}
        CB_IN_DEEPSEEK_BASE_URL: ${{ inputs.deepseek_base_url }}
        CB_IN_GLM_API_KEY: ${{ inputs.glm_api_key }}
        CB_IN_GLM_BASE_URL: ${{ inputs.glm_base_url }}
        CB_IN_GOOGLE_API_KEY: ${{ inputs.google_api_key }}
        CB_IN_KIMI_API_KEY: ${{ inputs.kimi_api_key }}
        CB_IN_KIMI_BASE_URL: ${{ inputs.kimi_base_url }}
        CB_IN_LITELLM_API_KEY: ${{ inputs.litellm_api_key }}
        CB_IN_LITELLM_BASE_URL: ${{ inputs.litellm_base_url }}
        CB_IN_OLLAMA_API_KEY: ${{ inputs.ollama_api_key }}
        CB_IN_OLLAMA_BASE_URL: ${{ inputs.ollama_base_url }}
        CB_IN_OPENAI_API_KEY: ${{ inputs.openai_api_key }}
        CB_IN_OPENAI_BASE_URL: ${{ inputs.openai_base_url }}
        CB_IN_OPENROUTER_API_KEY: ${{ inputs.openrouter_api_key }}
        CB_IN_OPPER_API_KEY: ${{ inputs.opper_api_key }}
        CB_IN_ORCAROUTER_API_KEY: ${{ inputs.orcarouter_api_key }}
        CB_IN_REQUESTY_API_KEY: ${{ inputs.requesty_api_key }}
        CB_IN_REQUESTY_BASE_URL: ${{ inputs.requesty_base_url }}
        CB_IN_VERCEL_API_KEY: ${{ inputs.vercel_api_key }}
        CB_IN_VERCEL_BASE_URL: ${{ inputs.vercel_base_url }}
      run: "$GITHUB_ACTION_PATH/scripts/action/verify-credentials.sh"

    # Said in the pull request, not only in the Actions tab, because the person who has
    # to add the secret is the one reading the pull request.
    - name: Report LLM configuration failure
      if: steps.guard.outputs.skip != 'true' && steps.llm.outputs.error != '' && steps.guard.outputs.mode == 'review' && steps.guard.outputs.pr_number != ''
      continue-on-error: true
      uses: marocchino/sticky-pull-request-comment@773744901bac0e8cbb5a0dc842800d45e9b2b405 # v2.9.4
      with:
        header: ${{ steps.guard.outputs.comment_id }}
        number: ${{ steps.guard.outputs.pr_number }}
        GITHUB_TOKEN: ${{ inputs.github_token }}
        message: |
          ### CodeBoarding review - not configured

          ${{ steps.llm.outputs.details }}

          No analysis ran, and no CodeBoarding hosted usage was consumed.

          <sub>run [${{ github.run_id }}](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}) - attempt ${{ github.run_attempt }}</sub>

    # Fires on a refusal AND on the check failing to reach a verdict at all. The step
    # above is `continue-on-error` so the refusal can be reported before the job dies,
    # which also means a crash in it -- an unreadable provider table, a missing python3 --
    # would leave `error` empty and let the run carry on to the checkout and the engine
    # install. Checking the outcome as well keeps "credentials are decided before anything
    # expensive happens" true when the deciding itself breaks.
    - name: Stop on LLM configuration failure
      if: steps.guard.outputs.skip != 'true' && (steps.llm.outputs.error != '' || steps.llm.outcome != 'success')
      shell: bash
      env:
        CB_ERROR: ${{ steps.llm.outputs.error }}
      run: |
        if [ -n "$CB_ERROR" ]; then
          echo "CodeBoarding did not run: $CB_ERROR" >&2
        else
          echo "::error title=CodeBoarding LLM configuration::The credential check did not complete, so no analysis was attempted. See the step above."
        fi
        exit 1

    # Before the checkout and the engine install, like the credential check: a sync
    # that could not save its result fails in seconds, not after a full analysis.
    - name: Check analysis branch
      id: analysis_check
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'sync'
      shell: bash
      env:
        BASELINE_BRANCH: ${{ steps.guard.outputs.analysis_branch }}
        SYNCED_BRANCH: ${{ steps.guard.outputs.synced_branch }}
        GIT_TOKEN: ${{ inputs.github_token }}
        GITHUB_SERVER_URL: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
      run: "$GITHUB_ACTION_PATH/scripts/action/check-analysis-branch.sh"

    - name: Update review progress
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      continue-on-error: true
      uses: marocchino/sticky-pull-request-comment@773744901bac0e8cbb5a0dc842800d45e9b2b405 # v2.9.4
      with:
        header: ${{ steps.guard.outputs.comment_id }}
        number: ${{ steps.guard.outputs.pr_number }}
        GITHUB_TOKEN: ${{ inputs.github_token }}
        message: |
          ### CodeBoarding review · analyzing…

          ⏳ CodeBoarding is analyzing this pull request's architecture changes.

          Open it in [CodeBoarding](https://app.codeboarding.org/${{ github.repository }}/pull/${{ steps.guard.outputs.pr_number }}?utm_source=github&utm_medium=pr_comment&utm_campaign=gh_action) meanwhile: the files, comments and review are there already, and the diff appears when the run finishes.

          <sub>run [${{ github.run_id }}](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}) · attempt ${{ github.run_attempt }}</sub>

    - name: Checkout analysis target
      if: steps.guard.outputs.skip != 'true'
      uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
      with:
        repository: ${{ steps.guard.outputs.checkout_repo }}
        ref: ${{ steps.guard.outputs.checkout_ref }}
        token: ${{ inputs.github_token }}
        fetch-depth: 1
        persist-credentials: false
        path: .codeboarding-target

    - name: Setup Java for CodeBoarding
      if: steps.guard.outputs.skip != 'true'
      uses: actions/setup-java@b6effb05e454b25005698d916606bdc6ffcbf961 # v5.7.0
      with:
        distribution: temurin
        java-version: '21'

    - name: Install CodeBoarding
      if: steps.guard.outputs.skip != 'true'
      shell: bash
      run: |
        python -m pip install --disable-pip-version-check 'codeboarding==0.15.3'
        # Fail here, with the reason, rather than mid-analysis with a traceback:
        # pinning a release does not pin what its dependencies resolve to. Run
        # the console script, not `python -c`, which would put the analyzed
        # repository's own main.py ahead of the CLI's on sys.path.
        codeboarding --help >/dev/null || {
          echo "::error::The installed CodeBoarding CLI cannot start; a dependency it imports is missing."
          exit 1
        }

    - name: Configure analysis authentication
      if: steps.guard.outputs.skip != 'true'
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
      run: "$GITHUB_ACTION_PATH/scripts/action/configure-auth.sh"

    - name: Resolve analysis cache keys
      id: state
      if: steps.guard.outputs.skip != 'true'
      shell: bash
      env:
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        MERGE_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        IS_FORK: ${{ steps.guard.outputs.is_fork }}
        LLM_PROVIDER: ${{ steps.llm.outputs.provider }}
        BACKEND_ID: ${{ steps.llm.outputs.backend_id }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
      run: "$GITHUB_ACTION_PATH/scripts/action/resolve-cache-keys.sh"

    # Both lookups are best effort. Without them the run derives the base from
    # the committed baseline and the head from the base, which is what every run
    # did before any of this existed.
    - name: Fetch published base analysis
      id: fetch_base
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.state.outputs.base_name != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      shell: bash
      env:
        GH_TOKEN: ${{ inputs.github_token }}
        GH_ENTERPRISE_TOKEN: ${{ inputs.github_token }}
        GH_HOST: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
        ARTIFACT_NAME: ${{ steps.state.outputs.base_name }}
        DEST: ${{ runner.temp }}/cb-state/${{ github.action }}/base
        # Must match the review retention below: a review references a base by
        # id for its whole life, so a base with less than that left is renewed.
        RENEW_WITHIN_DAYS: '14'
      run: "$GITHUB_ACTION_PATH/scripts/action/fetch-state.sh"

    - name: Fetch this pull request's last analysis
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.state.outputs.warmstart_name != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      shell: bash
      env:
        GH_TOKEN: ${{ inputs.github_token }}
        GH_ENTERPRISE_TOKEN: ${{ inputs.github_token }}
        GH_HOST: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
        ARTIFACT_NAME: ${{ steps.state.outputs.warmstart_name }}
        DEST: ${{ runner.temp }}/cb-state/${{ github.action }}/warmstart
      run: "$GITHUB_ACTION_PATH/scripts/action/fetch-state.sh"

    - name: Generate sync baseline
      id: sync_analyze
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'sync'
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        STAGE_DIR: ${{ runner.temp }}/cb-state/${{ github.action }}/out
        BASELINE_BRANCH: ${{ steps.guard.outputs.analysis_branch }}
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        # Lets a branch without a usable committed baseline catch up from a saved analysis.
        ANCESTOR_LOOKUP: ${{ github.server_url == 'https://github.com' && steps.state.outputs.cfg_hash != '' }}
        GIT_TOKEN: ${{ inputs.github_token }}
        REPOSITORY: ${{ github.repository }}
        GH_HOST: ${{ github.server_url }}
        GITHUB_SERVER_URL: ${{ github.server_url }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
      run: '"$GITHUB_ACTION_PATH/scripts/action/with-auth.sh" "$GITHUB_ACTION_PATH/scripts/action/sync-generate-baseline.sh"'

    - name: Deliver baseline
      id: sync_commit
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'sync'
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        ANALYSIS_DIR: ${{ steps.sync_analyze.outputs.analysis_dir }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        SYNCED_BRANCH: ${{ steps.guard.outputs.synced_branch }}
        BASELINE_BRANCH: ${{ steps.guard.outputs.analysis_branch }}
        ENGINE_VERSION: ${{ steps.state.outputs.engine_version }}
        # Recorded with each baseline-branch commit, so a reader can tell which configuration made it.
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        GITHUB_TOKEN: ${{ inputs.github_token }}
        GITHUB_SERVER_URL: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
      run: "$GITHUB_ACTION_PATH/scripts/action/deliver-sync.sh"

    # Named for the commit pull requests branch from, so a review that forks there
    # finds this graph and skips analyzing the base entirely. Published under both
    # the analyzed commit and the baseline commit written on top, because a pull
    # request opened either side of that commit has a different merge base.
    - name: Publish baseline analysis
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'sync' && steps.sync_commit.outputs.baseline_sha != '' && steps.state.outputs.cfg_hash != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: codeboarding-base-${{ steps.state.outputs.cfg_hash }}-${{ steps.sync_commit.outputs.baseline_sha }}
        path: ${{ runner.temp }}/cb-state/${{ github.action }}/out/base
        # Longer than a review's 14 days on purpose, and the margin is the whole
        # point: correctness comes from the renewal check, which republishes any
        # base with less than a review's lifetime left, while this surplus
        # decides how often that fires. Equal lifetimes renew on every run and
        # publish a base per run rather than per merge base — more storage, not
        # less. 30 leaves a 16-day reuse window.
        retention-days: 30
        if-no-files-found: ignore

    - name: Publish baseline analysis for the analyzed commit
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'sync' && steps.sync_commit.outputs.analyzed_sha != '' && steps.sync_commit.outputs.analyzed_sha != steps.sync_commit.outputs.baseline_sha && steps.state.outputs.cfg_hash != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: codeboarding-base-${{ steps.state.outputs.cfg_hash }}-${{ steps.sync_commit.outputs.analyzed_sha }}
        path: ${{ runner.temp }}/cb-state/${{ github.action }}/out/base
        # Longer than a review's 14 days on purpose, and the margin is the whole
        # point: correctness comes from the renewal check, which republishes any
        # base with less than a review's lifetime left, while this surplus
        # decides how often that fires. Equal lifetimes renew on every run and
        # publish a base per run rather than per merge base — more storage, not
        # less. 30 leaves a 16-day reuse window.
        retention-days: 30
        if-no-files-found: ignore

    - name: Write sync summary
      if: always() && steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'sync'
      shell: bash
      env:
        MODE: ${{ steps.sync_analyze.outputs.analysis_mode }}
        COMMITTED: ${{ steps.sync_commit.outputs.committed }}
        FILES: ${{ steps.sync_commit.outputs.files_written }}
        SAVED_TO: ${{ steps.guard.outputs.analysis_branch || steps.guard.outputs.synced_branch }}
        # A failed sync has no pull request to comment on; the summary is where it says why.
        FAILURE_REASON: ${{ steps.guard.outputs.failure_reason || steps.analysis_check.outputs.failure_reason || steps.state.outputs.failure_reason || steps.sync_analyze.outputs.failure_reason || steps.sync_commit.outputs.failure_reason }}
      run: "$GITHUB_ACTION_PATH/scripts/action/sync-summary.sh"

    - name: Generate review baseline
      id: review_baseline
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        REVIEW_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        REVIEW_BASE_REPO: ${{ steps.guard.outputs.base_repo }}
        BASE_DIR: ${{ runner.temp }}/cb-state/${{ github.action }}/base
        RENEW_BASE: ${{ steps.fetch_base.outputs.renew }}
        BASELINE_BRANCH: ${{ steps.guard.outputs.analysis_branch }}
        # Lets a merge base without a saved analysis catch up from an ancestor's.
        ANCESTOR_LOOKUP: ${{ github.server_url == 'https://github.com' && steps.state.outputs.cfg_hash != '' }}
        # For the progress comment while a base is built from scratch.
        PROGRESS_HEADER: ${{ steps.guard.outputs.comment_id }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        BASE_REF: ${{ steps.guard.outputs.base_ref }}
        REPOSITORY: ${{ github.repository }}
        GH_HOST: ${{ github.server_url }}
        STAGE_DIR: ${{ runner.temp }}/cb-state/${{ github.action }}/out
        ENGINE_VERSION: ${{ steps.state.outputs.engine_version }}
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        GIT_TOKEN: ${{ inputs.github_token }}
        GITHUB_SERVER_URL: ${{ github.server_url }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
        # The head step below analyzes with the same credentials, and removes them.
        KEEP_AUTH: 'true'
      run: '"$GITHUB_ACTION_PATH/scripts/action/with-auth.sh" "$GITHUB_ACTION_PATH/scripts/action/review-generate-baseline.sh"'

    - name: Analyze pull request
      id: review_analyze
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        BASE_ANALYSIS_PATH: ${{ steps.review_baseline.outputs.base_analysis_path }}
        WARMSTART_DIR: ${{ runner.temp }}/cb-state/${{ github.action }}/warmstart
        REVIEW_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        REVIEW_HEAD_SHA: ${{ steps.guard.outputs.head_sha }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        STAGE_DIR: ${{ runner.temp }}/cb-state/${{ github.action }}/out
        ENGINE_VERSION: ${{ steps.state.outputs.engine_version }}
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
        # A setup PR's run goes on to preview other pull requests; the credentials stay until
        # those finish and the step after them removes them.
        KEEP_AUTH: ${{ github.event_name == 'pull_request' && github.head_ref == 'codeboarding/setup-action' }}
      run: '"$GITHUB_ACTION_PATH/scripts/action/with-auth.sh" "$GITHUB_ACTION_PATH/scripts/action/review-analyze-head.sh"'

    # Published before the review is rendered: the analysis is the expensive part,
    # so a later rendering or posting failure must not throw it away. Only the
    # next run reads this, hence the short retention.
    - name: Publish this analysis for the next run
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.state.outputs.warmstart_name != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: ${{ steps.state.outputs.warmstart_name }}
        path: ${{ runner.temp }}/cb-state/${{ github.action }}/out/warmstart
        retention-days: ${{ inputs.warmstart_retention_days }}
        if-no-files-found: ignore

    # Only a trusted run publishes a base graph: it is named for a commit, so
    # every pull request forking there would read it, and a fork's analysis must
    # never be what they read.
    - name: Publish base analysis
      id: publish_base
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.review_baseline.outputs.publish_base == 'true' && steps.guard.outputs.is_fork != 'true' && steps.state.outputs.base_name != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: ${{ steps.state.outputs.base_name }}
        path: ${{ runner.temp }}/cb-state/${{ github.action }}/out/base
        # Longer than a review's 14 days on purpose, and the margin is the whole
        # point: correctness comes from the renewal check, which republishes any
        # base with less than a review's lifetime left, while this surplus
        # decides how often that fires. Equal lifetimes renew on every run and
        # publish a base per run rather than per merge base — more storage, not
        # less. 30 leaves a 16-day reuse window.
        retention-days: 30
        if-no-files-found: ignore

    - name: Render review diagram
      id: review_render
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        BASE_ANALYSIS_PATH: ${{ steps.review_baseline.outputs.base_analysis_path }}
        HEAD_ANALYSIS_PATH: ${{ steps.review_analyze.outputs.analysis_path }}
      run: "$GITHUB_ACTION_PATH/scripts/action/render-review.sh"

    - name: Build review artifact
      id: review_artifact
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      shell: bash
      env:
        ANALYSIS_PATH: ${{ steps.review_analyze.outputs.analysis_path }}
        BASE_ARTIFACT_NAME: ${{ steps.state.outputs.base_name }}
        BASE_ARTIFACT_ID: ${{ steps.publish_base.outputs.artifact-id || steps.fetch_base.outputs.artifact_id }}
        BASE_ANALYSIS_PATH: ${{ steps.review_baseline.outputs.base_analysis_path }}
        INLINE_BASE: ${{ steps.publish_base.outputs.artifact-id == '' && steps.fetch_base.outputs.artifact_id == '' }}
        ANALYSIS_MODE: ${{ steps.review_analyze.outputs.analysis_mode }}
        BASE_SHA: ${{ steps.guard.outputs.base_sha }}
        MERGE_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        MERGE_BASE_RESOLVED: ${{ steps.guard.outputs.merge_base_resolved }}
        HEAD_SHA: ${{ steps.guard.outputs.head_sha }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        SEED_SOURCE: ${{ steps.review_analyze.outputs.seed_source }}
        CHAIN_DEPTH: ${{ steps.review_analyze.outputs.chain_depth }}
        ANALYSED_FILES_CHANGED: ${{ steps.review_render.outputs.analysed_files_changed }}
      run: "$GITHUB_ACTION_PATH/scripts/action/build-review-artifact.sh"

    - name: Upload review artifact
      id: upload_review_artifact_dotcom
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && github.server_url == 'https://github.com'
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: codeboarding-review-${{ github.run_id }}-${{ github.run_attempt }}
        path: ${{ steps.review_artifact.outputs.artifact_dir }}
        if-no-files-found: error
        # The dominant storage cost, since it is the one kept for weeks. A pull
        # request open longer than this loses its rendered analysis until someone
        # asks for it again, which costs one incremental over the pull request:
        # the base graph is still published and needs no re-analysis.
        retention-days: 14

    - name: Build review comment
      id: review_body
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      shell: bash
      env:
        DIAGRAM: ${{ steps.review_render.outputs.diagram_md }}
        N_CHANGED: ${{ steps.review_render.outputs.n_changed }}
        ARTIFACT_URL: ${{ steps.upload_review_artifact_dotcom.outputs.artifact-url }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        BEHIND_BY: ${{ steps.guard.outputs.behind_by }}
        BASE_REF: ${{ steps.guard.outputs.base_ref }}
        MERGE_BASE_RESOLVED: ${{ steps.guard.outputs.merge_base_resolved }}
        HEAD_SHA: ${{ steps.guard.outputs.head_sha }}
        # Analysed files whose content hash differs between base and head, from the render step.
        ANALYSED_FILES_CHANGED: ${{ steps.review_render.outputs.analysed_files_changed }}
      run: "$GITHUB_ACTION_PATH/scripts/action/build-review-comment.sh"

    - name: Post review comment
      id: review_comment
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      uses: marocchino/sticky-pull-request-comment@773744901bac0e8cbb5a0dc842800d45e9b2b405 # v2.9.4
      with:
        header: ${{ steps.guard.outputs.comment_id }}
        number: ${{ steps.guard.outputs.pr_number }}
        GITHUB_TOKEN: ${{ inputs.github_token }}
        path: ${{ steps.review_body.outputs.path }}

    # Before a setup PR merges, its own run is the only one GitHub starts for us, so this is
    # where a new user's open pull requests get their first review: the ones the app listed
    # in the setup PR's description, each applied to the commit this run just analysed and
    # analysed incrementally from it (see review-preview-pr.sh). Nothing is posted on them.
    # The app finds each as codeboarding-preview-<run>-<attempt>-pr<number>-<head sha>.
    # Not codeboarding-review-<run>-...: the app takes the newest artifact with that prefix as
    # the setup PR's own review, so a preview named that way would replace the setup PR's map.
    # All best effort: the setup PR's map is already uploaded and must not fail because of them.
    - name: Resolve pull requests to preview
      id: preview_list
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.review_comment.outcome == 'success' && github.event_name == 'pull_request' && github.head_ref == 'codeboarding/setup-action'
      continue-on-error: true
      shell: bash
      env:
        GH_TOKEN: ${{ inputs.github_token }}
        GH_HOST: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        HEAD_REF: ${{ github.head_ref }}
        BASE_REF: ${{ steps.guard.outputs.base_ref }}
        MAX_PREVIEWS: '3'
      run: "$GITHUB_ACTION_PATH/scripts/action/resolve-preview-prs.sh"

    - name: Preview pull request 1 for the setup PR
      id: preview_1
      if: steps.guard.outputs.mode == 'review' && steps.preview_list.outputs.count >= 1
      continue-on-error: true
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        GIT_TOKEN: ${{ inputs.github_token }}
        REPOSITORY: ${{ github.repository }}
        SETUP_PR: ${{ steps.guard.outputs.pr_number }}
        REVIEW_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        BASE_ANALYSIS_PATH: ${{ steps.review_baseline.outputs.base_analysis_path }}
        # The setup PR's own review artifact already records where the base graph lives.
        REVIEW_ARTIFACT_DIR: ${{ steps.review_artifact.outputs.artifact_dir }}
        ENGINE_VERSION: ${{ steps.state.outputs.engine_version }}
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
        KEEP_AUTH: 'true'
      run: '"$GITHUB_ACTION_PATH/scripts/action/with-auth.sh" "$GITHUB_ACTION_PATH/scripts/action/review-preview-pr.sh" 1'

    - name: Upload preview 1
      id: upload_preview_1
      if: steps.preview_1.outputs.artifact_dir != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: codeboarding-preview-${{ github.run_id }}-${{ github.run_attempt }}-pr${{ steps.preview_1.outputs.pr_number }}-${{ steps.preview_1.outputs.head_sha }}
        path: ${{ steps.preview_1.outputs.artifact_dir }}
        if-no-files-found: error
        retention-days: 14

    # The sticky review comment is an edit by now, and edits send no notification. One new
    # comment on a setup PR emails its author: after the first preview when there is one,
    # so the email links a coloured pull request, otherwise as soon as the map is posted.
    - name: Tell the setup PR's author the map is ready
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.review_comment.outcome == 'success' && github.event_name == 'pull_request' && github.head_ref == 'codeboarding/setup-action'
      continue-on-error: true
      shell: bash
      env:
        GH_TOKEN: ${{ inputs.github_token }}
        GH_HOST: ${{ github.server_url }}
        REPOSITORY: ${{ github.repository }}
        PR_NUMBER: ${{ steps.guard.outputs.pr_number }}
        PR_AUTHOR: ${{ github.event.pull_request.user.login }}
        HEAD_REF: ${{ github.head_ref }}
        BASE_REF: ${{ steps.guard.outputs.base_ref }}
        FIRST_PR: ${{ steps.upload_preview_1.outcome == 'success' && steps.preview_1.outputs.pr_number || '' }}
        FIRST_TITLE: ${{ steps.preview_1.outputs.title }}
        PREVIEW_PRS: ${{ steps.preview_list.outputs.prs }}
      run: "$GITHUB_ACTION_PATH/scripts/action/notify-setup-ready.sh"

    - name: Preview pull request 2 for the setup PR
      id: preview_2
      if: steps.guard.outputs.mode == 'review' && steps.preview_list.outputs.count >= 2
      continue-on-error: true
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        GIT_TOKEN: ${{ inputs.github_token }}
        REPOSITORY: ${{ github.repository }}
        SETUP_PR: ${{ steps.guard.outputs.pr_number }}
        REVIEW_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        BASE_ANALYSIS_PATH: ${{ steps.review_baseline.outputs.base_analysis_path }}
        # The setup PR's own review artifact already records where the base graph lives.
        REVIEW_ARTIFACT_DIR: ${{ steps.review_artifact.outputs.artifact_dir }}
        ENGINE_VERSION: ${{ steps.state.outputs.engine_version }}
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
        KEEP_AUTH: 'true'
      run: '"$GITHUB_ACTION_PATH/scripts/action/with-auth.sh" "$GITHUB_ACTION_PATH/scripts/action/review-preview-pr.sh" 2'

    - name: Upload preview 2
      id: upload_preview_2
      if: steps.preview_2.outputs.artifact_dir != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: codeboarding-preview-${{ github.run_id }}-${{ github.run_attempt }}-pr${{ steps.preview_2.outputs.pr_number }}-${{ steps.preview_2.outputs.head_sha }}
        path: ${{ steps.preview_2.outputs.artifact_dir }}
        if-no-files-found: error
        retention-days: 14

    - name: Preview pull request 3 for the setup PR
      id: preview_3
      if: steps.guard.outputs.mode == 'review' && steps.preview_list.outputs.count >= 3
      continue-on-error: true
      shell: bash
      env:
        ACTION_PATH: ${{ github.action_path }}
        CHECKOUT_DIR: ${{ github.workspace }}/.codeboarding-target
        GIT_TOKEN: ${{ inputs.github_token }}
        REPOSITORY: ${{ github.repository }}
        SETUP_PR: ${{ steps.guard.outputs.pr_number }}
        REVIEW_BASE_SHA: ${{ steps.guard.outputs.merge_base_sha }}
        BASE_ANALYSIS_PATH: ${{ steps.review_baseline.outputs.base_analysis_path }}
        # The setup PR's own review artifact already records where the base graph lives.
        REVIEW_ARTIFACT_DIR: ${{ steps.review_artifact.outputs.artifact_dir }}
        ENGINE_VERSION: ${{ steps.state.outputs.engine_version }}
        CFG_HASH: ${{ steps.state.outputs.cfg_hash }}
        DEPTH_CAP: ${{ inputs.depth_cap }}
        MODEL: ${{ inputs.model }}
        AGENT_MODEL_INPUT: ${{ inputs.agent_model }}
        PARSING_MODEL_INPUT: ${{ inputs.parsing_model }}
      run: '"$GITHUB_ACTION_PATH/scripts/action/with-auth.sh" "$GITHUB_ACTION_PATH/scripts/action/review-preview-pr.sh" 3'

    - name: Upload preview 3
      id: upload_preview_3
      if: steps.preview_3.outputs.artifact_dir != '' && github.server_url == 'https://github.com'
      continue-on-error: true
      uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
      with:
        name: codeboarding-preview-${{ github.run_id }}-${{ github.run_attempt }}-pr${{ steps.preview_3.outputs.pr_number }}-${{ steps.preview_3.outputs.head_sha }}
        path: ${{ steps.preview_3.outputs.artifact_dir }}
        if-no-files-found: error
        retention-days: 14

    # The head and preview steps remove the credentials themselves; this covers a run that
    # failed before them, and a setup PR's run, which keeps them for its previews.
    - name: Remove analysis credentials
      if: always() && steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review'
      shell: bash
      run: '"$GITHUB_ACTION_PATH/scripts/action/clear-auth.sh"'

    # The same body, on the run page. Job summaries and logs are explicitly excluded from
    # the artifact storage allowance, and the cap is 1MiB per step against a diagram
    # measured in single-digit kilobytes, so this is free in both senses. It earns its
    # place on the triggers where the comment cannot be the record: a manual dispatch, or
    # any run whose comment write fails on a token without `pull-requests: write`.
    - name: Add the review to the job summary
      if: steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.review_body.outputs.path != ''
      continue-on-error: true
      shell: bash
      env:
        BODY: ${{ steps.review_body.outputs.path }}
      run: cat "$BODY" >> "$GITHUB_STEP_SUMMARY"

    # Skipped when the run stopped on a credential problem: that path already replaced
    # this same sticky comment with the input and secret to fix, and "see the workflow
    # logs" posted over the top of it would send the reader hunting for what they had
    # just been told.
    - name: Post review failure
      if: failure() && steps.guard.outputs.skip != 'true' && steps.guard.outputs.mode == 'review' && steps.review_comment.outcome != 'success' && steps.llm.outputs.error == ''
      continue-on-error: true
      uses: marocchino/sticky-pull-request-comment@773744901bac0e8cbb5a0dc842800d45e9b2b405 # v2.9.4
      with:
        header: ${{ steps.guard.outputs.comment_id }}
        number: ${{ steps.guard.outputs.pr_number }}
        GITHUB_TOKEN: ${{ inputs.github_token }}
        message: |
          ### CodeBoarding review · failed

          ${{ steps.review_baseline.outputs.failure_reason || steps.review_analyze.outputs.failure_reason }}

          See the [workflow logs](${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}).