action.yml

name: "deltatau-audit"
description: "Time robustness audit for RL agents — find timing failures before deployment"
branding:
  icon: "shield"
  color: "blue"

inputs:
  command:
    description: "Command to run: demo, audit-sb3, audit-cleanrl, audit-hf, fix-sb3"
    required: false
    default: "demo"
  model:
    description: "Path to SB3 model (.zip) — for audit-sb3, audit-hf, fix-sb3"
    required: false
  algo:
    description: "SB3 algorithm (ppo, sac, td3, a2c)"
    required: false
    default: "ppo"
  checkpoint:
    description: "Path to agent checkpoint (.pt file) — for audit-cleanrl, fix-cleanrl"
    required: false
  agent-module:
    description: "Python module path for CleanRL agent class (e.g. 'my_pkg.agent:Agent') — for audit-cleanrl"
    required: false
  env:
    description: "Gymnasium environment ID (e.g. CartPole-v1, HalfCheetah-v5)"
    required: false
  episodes:
    description: "Episodes per condition"
    required: false
    default: "20"
  output:
    description: "Output directory"
    required: false
    default: "audit_report"
  extras:
    description: "pip extras to install (e.g. 'sb3,mujoco')"
    required: false
    default: "demo"
  deploy-threshold:
    description: "Deployment score threshold for pass (0-1)"
    required: false
    default: "0.80"
  stress-threshold:
    description: "Stress score threshold for warn (0-1)"
    required: false
    default: "0.50"
  workers:
    description: "Parallel workers for episode collection (integer or 'auto')"
    required: false
    default: "auto"
  seed:
    description: "Random seed for reproducible results"
    required: false
  seeds:
    description: "Multi-seed sweep list (space/comma separated, e.g. '0 1 2 3 4')"
    required: false
    default: ""
  ci-min-deployment-pass-rate:
    description: "Multi-seed gate: minimum deployment pass rate (0-1)"
    required: false
    default: "0.80"
  ci-min-stress-pass-rate:
    description: "Multi-seed gate: minimum stress pass rate (0-1)"
    required: false
    default: "0.50"
  extra-args:
    description: "Additional CLI arguments (e.g. '--adaptive --target-ci-width 0.05')"
    required: false
    default: ""
  python-version:
    description: "Python version"
    required: false
    default: "3.11"

outputs:
  status:
    description: "Audit status: pass, warn, or fail"
    value: ${{ steps.audit.outputs.status }}
  deployment-score:
    description: "Deployment robustness score (0-1)"
    value: ${{ steps.audit.outputs.deployment_score }}
  stress-score:
    description: "Stress robustness score (0-1)"
    value: ${{ steps.audit.outputs.stress_score }}
  report-dir:
    description: "Path to the generated report directory"
    value: ${{ inputs.output }}

runs:
  using: "composite"
  steps:
    - name: Set up Python
      uses: actions/setup-python@v5
      with:
        python-version: ${{ inputs.python-version }}

    - name: Cache pip
      uses: actions/cache@v4
      with:
        path: ~/.cache/pip
        key: ${{ runner.os }}-pip-deltatau-${{ inputs.extras }}-${{ hashFiles('**/pyproject.toml') }}
        restore-keys: |
          ${{ runner.os }}-pip-deltatau-${{ inputs.extras }}-
          ${{ runner.os }}-pip-deltatau-

    - name: Install deltatau-audit
      shell: bash
      run: pip install "deltatau-audit[${{ inputs.extras }}]"

    - name: Run audit
      id: audit
      shell: bash
      run: |
        set +e

        SINGLE_SEED_FLAG=""
        if [ -n "${{ inputs.seed }}" ]; then
          SINGLE_SEED_FLAG="--seed ${{ inputs.seed }}"
        fi

        SWEEP_SEEDS_FLAG=""
        if [ -n "${{ inputs.seeds }}" ]; then
          CLEAN_SEEDS=$(echo "${{ inputs.seeds }}" | tr ',' ' ')
          SWEEP_SEEDS_FLAG="--seeds $CLEAN_SEEDS"
        fi

        AUDIT_SEED_FLAGS="$SINGLE_SEED_FLAG"
        if [ -n "$SWEEP_SEEDS_FLAG" ]; then
          AUDIT_SEED_FLAGS="$SWEEP_SEEDS_FLAG"
        fi

        COMMON_FLAGS="--ci \
          --episodes ${{ inputs.episodes }} \
          --workers ${{ inputs.workers }} \
          --ci-deploy-threshold ${{ inputs.deploy-threshold }} \
          --ci-stress-threshold ${{ inputs.stress-threshold }} \
          --ci-min-deployment-pass-rate ${{ inputs.ci-min-deployment-pass-rate }} \
          --ci-min-stress-pass-rate ${{ inputs.ci-min-stress-pass-rate }} \
          --format markdown \
          --out ${{ inputs.output }} \
          ${{ inputs.extra-args }}"

        if [ "${{ inputs.command }}" = "demo" ]; then
          python -m deltatau_audit demo cartpole $COMMON_FLAGS $SINGLE_SEED_FLAG
          EXIT_CODE=$?

        elif [ "${{ inputs.command }}" = "audit-sb3" ]; then
          python -m deltatau_audit audit-sb3 \
            --algo ${{ inputs.algo }} \
            --model "${{ inputs.model }}" \
            --env "${{ inputs.env }}" \
            $COMMON_FLAGS $AUDIT_SEED_FLAGS
          EXIT_CODE=$?

        elif [ "${{ inputs.command }}" = "audit-cleanrl" ]; then
          CHECKPOINT_FLAG=""
          if [ -n "${{ inputs.checkpoint }}" ]; then
            CHECKPOINT_FLAG="--checkpoint ${{ inputs.checkpoint }}"
          fi
          AGENT_MODULE_FLAG=""
          if [ -n "${{ inputs.agent-module }}" ]; then
            AGENT_MODULE_FLAG="--agent-module ${{ inputs.agent-module }}"
          fi
          python -m deltatau_audit audit-cleanrl \
            --env "${{ inputs.env }}" \
            $CHECKPOINT_FLAG \
            $AGENT_MODULE_FLAG \
            $COMMON_FLAGS $AUDIT_SEED_FLAGS
          EXIT_CODE=$?

        elif [ "${{ inputs.command }}" = "audit-hf" ]; then
          python -m deltatau_audit audit-hf \
            --algo ${{ inputs.algo }} \
            --repo "${{ inputs.model }}" \
            --env "${{ inputs.env }}" \
            $COMMON_FLAGS $AUDIT_SEED_FLAGS
          EXIT_CODE=$?

        elif [ "${{ inputs.command }}" = "fix-sb3" ]; then
          python -m deltatau_audit fix-sb3 \
            --algo ${{ inputs.algo }} \
            --model "${{ inputs.model }}" \
            --env "${{ inputs.env }}" \
            $COMMON_FLAGS $SINGLE_SEED_FLAG
          EXIT_CODE=$?

        else
          echo "Unknown command: ${{ inputs.command }}"
          exit 1
        fi

        # Parse CI summary for outputs
        if [ "${{ inputs.command }}" = "demo" ]; then
          CI_JSON="${{ inputs.output }}/robust_wide/ci_summary.json"
        else
          CI_JSON="${{ inputs.output }}/ci_summary.json"
        fi

        if [ -f "$CI_JSON" ]; then
          read -r STATUS DEP STR <<< $(python -c "import json; d=json.load(open('$CI_JSON')); print(d['status'], d.get('deployment_score', 0), d.get('stress_score', 0))")
          echo "status=$STATUS" >> $GITHUB_OUTPUT
          echo "deployment_score=$DEP" >> $GITHUB_OUTPUT
          echo "stress_score=$STR" >> $GITHUB_OUTPUT
        else
          echo "status=unknown" >> $GITHUB_OUTPUT
          echo "deployment_score=0" >> $GITHUB_OUTPUT
          echo "stress_score=0" >> $GITHUB_OUTPUT
        fi

        exit $EXIT_CODE