story tool

.env.example

@@ -1,6 +1,10 @@
 # GitHub token for API access (required)
 GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxx
 
+# Optional: when the prr install has no .git/, stamp startup/output.log with this commit (short or full SHA).
+# If .git exists in the prr package root, revision is taken from git instead. Do not use another repo's GITHUB_SHA.
+# PRR_GIT_SHA=abc1234
+
 # ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 # ElizaCloud - Unified Model Gateway (RECOMMENDED)
 # ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
@@ -14,6 +18,12 @@ ELIZACLOUD_API_KEY=your-elizacloud-key-here
 # Default model (optional, defaults to gpt-4o):
 # PRR_LLM_MODEL=gpt-4o
 
+# ElizaCloud: extra 500/502/504 retries inside each complete() (0–15). CI defaults to 5 HTTP attempts when unset; locally 3.
+# PRR_ELIZACLOUD_SERVER_ERROR_RETRIES=6
+
+# After 2 consecutive gateway 5xx on the same model, try these ids in order (comma-separated). off = disable.
+# PRR_ELIZACLOUD_GATEWAY_FALLBACK_MODELS=anthropic/claude-3-5-haiku-20241022,openai/gpt-4o-mini
+
 # ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 # Alternative: Direct Provider Keys
 # ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
@@ -32,5 +42,66 @@ ELIZACLOUD_API_KEY=your-elizacloud-key-here
 # RECOMMENDED: Leave commented out for auto-detect.
 # PRR_TOOL=elizacloud
 
-# Extended thinking token budget for Claude models (optional)
+# Extended thinking token budget for Claude models (optional). Values above 500,000 clamp with a warning.
 # PRR_THINKING_BUDGET=10000
+
+# LLM concurrency (default 1). Raised values speed batches when the gateway allows it.
+# PRR_MAX_CONCURRENT_LLM=3
+
+# Optional: max ms per concurrent pool task (batch analysis / parallel fix groups). Unset or 0 = no cap.
+# PRR_LLM_TASK_TIMEOUT_MS=600000
+
+# Skip a fixer model for the rest of the run after this many verification failures with zero verified fixes (default 4). Set to 0 to disable.
+# PRR_SESSION_MODEL_SKIP_FAILURES=4
+# Every N fix iterations, clear session-skipped models so rotation retries them (0 = off). Pill-output #847.
+# PRR_SESSION_MODEL_SKIP_RESET_AFTER_FIX_ITERATIONS=8
+
+# Warn once after this many consecutive iterations with no new verified fixes (default 10). Set to 0 to disable.
+# PRR_DIMINISHING_RETURNS_ITERATIONS=10
+
+# ElizaCloud: built-in skip list is in shared/constants.ts (ELIZACLOUD_SKIP_MODEL_IDS).
+# Add more models to skip for this machine/run (comma-separated API ids, merged with built-in list):
+# PRR_ELIZACLOUD_EXTRA_SKIP_MODELS=openai/some-model-id
+
+# ElizaCloud: re-enable specific models removed from the default skip list (comma-separated API ids).
+# PRR_ELIZACLOUD_INCLUDE_MODELS=openai/gpt-4o
+
+# Merge conflict preprocessing: disable automatic insertion of missing ======= between markers.
+# PRR_DISABLE_CONFLICT_SEPARATOR_REPAIR=1
+
+# Catalog model advice (see AGENTS.md): disable 0a6 dismissal and/or quoted-literal auto-heal.
+# PRR_DISABLE_MODEL_CATALOG_SOLVABILITY=1
+# PRR_DISABLE_MODEL_CATALOG_AUTOHEAL=1
+
+# Thread replies: set to the GitHub login that posts replies so re-runs skip duplicate posts.
+# PRR_BOT_LOGIN=my-bot
+# PRR_REPLY_TO_THREADS=true
+# Also reply on threads dismissed as chronic-failure (default: no — batch token-saving dismissals).
+# PRR_THREAD_REPLY_INCLUDE_CHRONIC_FAILURE=1
+
+# Exit 2 on success when final audit passed any issue via UNCERTAIN or truncation guard (fail closed).
+# PRR_STRICT_FINAL_AUDIT_UNCERTAIN=1
+
+# Exit setup before clone when CodeRabbit (or bot) review SHA ≠ PR HEAD — saves work if inline comments are stale.
+# PRR_EXIT_ON_STALE_BOT_REVIEW=1
+
+# Exit setup before clone when GitHub says PR is not mergeable / dirty (unless you pass --merge-base).
+# PRR_EXIT_ON_UNMERGEABLE=1
+
+# On PR HEAD change: clear every dismissal (default clears only "already-fixed" dismissals).
+# PRR_CLEAR_ALL_DISMISSED_ON_HEAD=1
+
+# Dry-merge probes after fetch (git merge-tree; warns before pull if conflicts):
+# PRR_DISABLE_LATENT_MERGE_PROBE=1          # skip HEAD vs origin/<prBranch>
+# PRR_DISABLE_LATENT_MERGE_PROBE_BASE=1     # skip HEAD vs origin/<prBase> (GitHub mergeable/dirty)
+
+# Materialize a probe hit so LLM conflict resolution runs before pull:
+# PRR_MATERIALIZE_LATENT_MERGE=1            # merge origin/<prBranch> --no-commit
+# PRR_MATERIALIZE_LATENT_MERGE_BASE=1       # merge origin/<prBase> --no-commit
+
+# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+# Pill (log audit — optional, e.g. prr --pill)
+# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+# When tools/prr exists under the pill target directory, drop improvements whose file paths
+# look like the PR clone (src/, packages/, …). Set to 0 to record everything the LLM returns.
+# PILL_TOOL_REPO_SCOPE_FILTER=0

.github/workflows/eval.yml

@@ -0,0 +1,68 @@
+name: Run Evals
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+  schedule:
+    # Run full evals daily
+    - cron: '0 2 * * *'
+  workflow_dispatch:
+
+jobs:
+  eval:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        tool: [prr, pill, split-plan, story]
+      fail-fast: false
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '18'
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Compile (emit dist/)
+        run: npm run build
+
+      - name: Run evals
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          ELIZACLOUD_API_KEY: ${{ secrets.ELIZACLOUD_API_KEY }}
+        run: |
+          # Run fast evals on PRs, full evals on main/schedule
+          if [ "${{ github.event_name }}" = "pull_request" ]; then
+            echo "Running fast evals (subset) for PR"
+            npm run eval -- run -t ${{ matrix.tool }} -b simple-fix || true
+          else
+            echo "Running full evals"
+            npm run eval -- run -t ${{ matrix.tool }} -b simple-fix --compare || true
+            # Add more benchmarks as they're created
+          fi
+        continue-on-error: true
+
+      - name: Upload eval results
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: eval-results-${{ matrix.tool }}
+          path: tests/evals/results/
+          retention-days: 30
+
+      - name: Check for regressions
+        if: github.event_name != 'pull_request'
+        run: |
+          # TODO: Implement regression check
+          # Fail if metrics regress significantly
+          echo "Regression check not yet implemented"

.github/workflows/lint-workflows.yml

@@ -0,0 +1,19 @@
+# Validate workflow syntax and expressions before they run.
+# Catches: unrecognized context (e.g. env in job if), invalid expressions, bad refs.
+# Run locally: npm run lint:workflows (requires actionlint in PATH, e.g. brew install actionlint).
+name: Lint workflows
+
+on:
+  push:
+    paths: ['.github/workflows/**']
+  pull_request:
+    paths: ['.github/workflows/**']
+
+jobs:
+  actionlint:
+    name: actionlint
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Run actionlint
+        uses: rhysd/actionlint@v1.7.11

.github/workflows/refresh-model-catalog.yml

@@ -0,0 +1,38 @@
+# Refreshes generated/model-provider-catalog.json from official Anthropic + OpenAI doc URLs.
+# WHY: Vendor model IDs change; a committed snapshot keeps PRR/tools aligned without manual edits.
+name: Refresh model catalog
+
+on:
+  schedule:
+    - cron: '0 6 * * 1' # Mondays 06:00 UTC (~weekly)
+  workflow_dispatch:
+
+permissions:
+  contents: write
+
+jobs:
+  update:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+          cache: npm
+
+      - run: npm ci
+
+      - run: npm run update-model-catalog
+
+      - name: Commit and push if changed
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git add generated/model-provider-catalog.json
+          if git diff --staged --quiet; then
+            echo "No catalog changes."
+          else
+            git commit -m "chore: refresh model-provider-catalog.json"
+            git push
+          fi

.github/workflows/run-prr-client.yml

@@ -4,10 +4,11 @@
 #   1. Actions → Run PRR (client) → Run workflow → enter PR number.
 #   2. On the PR: add the "run-prr" label.
 #   3. Request PRR as a reviewer: add a bot user (or GitHub App bot) as a collaborator,
-#      set PRR_REVIEWER_LOGIN to that bot's login (e.g. prr-bot or my-app[bot]), then
-#      when someone requests that reviewer on the PR, this workflow runs.
+#      set repository variable PRR_REVIEWER_LOGIN (Settings → Variables) to that bot's
+#      login (e.g. prr-bot or my-app[bot]); when that reviewer is requested, this workflow runs.
 #
-# WHY pass secrets.GITHUB_TOKEN: The job runs in this repo's context; no extra PAT needed.
+# Inlined from run-prr-server.yml so validators that require runs-on accept this file.
+# Other repos can still call run-prr-server.yml directly (workflow_call).
 name: Run PRR (client)
 
 on:
@@ -26,23 +27,71 @@ concurrency:
 
 jobs:
   prr:
-    # Run when: manual, or label "run-prr", or requested reviewer matches PRR_REVIEWER_LOGIN (set in env below).
+    name: Run PRR
+    # Run when: manual, or label "run-prr", or requested reviewer matches vars.PRR_REVIEWER_LOGIN.
+    # (env is not available in job-level if; use repository variable instead.)
     if: |
       github.event_name == 'workflow_dispatch' ||
       (github.event_name == 'pull_request' && (
         (github.event.action == 'labeled' && github.event.label.name == 'run-prr') ||
-        (github.event.action == 'review_requested' && env.PRR_REVIEWER_LOGIN != '' && github.event.requested_reviewer.login == env.PRR_REVIEWER_LOGIN)
+        (github.event.action == 'review_requested' && vars.PRR_REVIEWER_LOGIN != '' && github.event.requested_reviewer.login == vars.PRR_REVIEWER_LOGIN)
       ))
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
     env:
-      # Set to your bot's login (e.g. prr-bot or my-app[bot]) to run when that user is requested as a reviewer. Leave empty to disable.
-      PRR_REVIEWER_LOGIN: ''
-    uses: ./.github/workflows/run-prr-server.yml
-    with:
-      pr_number: ${{ github.event_name == 'workflow_dispatch' && inputs.pr_number || github.event.pull_request.number }}
-      prr_repo: ${{ github.repository }}
-    secrets:
-      # Caller's token; server receives it as PRR_GITHUB_TOKEN and uses it as GITHUB_TOKEN.
-      PRR_GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-      ELIZACLOUD_API_KEY: ${{ secrets.ELIZACLOUD_API_KEY }}
-      ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-      OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
+    permissions:
+      contents: write      # checkout + PRR push
+      pull-requests: write # submit PR review (Create a review for a pull request)
+    steps:
+      - name: Checkout PRR
+        uses: actions/checkout@v4
+
+      - name: Resolve PR number
+        if: github.event_name == 'workflow_dispatch'
+        id: resolve-pr
+        run: echo "pr_number=${{ inputs.pr_number }}" >> $GITHUB_OUTPUT
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '24'
+          cache: 'npm'
+          cache-dependency-path: package-lock.json
+
+      # dist/ is gitignored — must compile before node dist/tools/prr/index.js (not a redundant typecheck).
+      - name: Install and compile PRR
+        run: |
+          npm ci
+          npm run build
+
+      - name: Configure Git
+        run: |
+          git config --global user.name "github-actions[bot]"
+          git config --global user.email "41898282+github-actions[bot]@users.noreply.github.com"
+
+      - name: Run PRR
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          PRR_SUBMIT_REVIEW: ${{ github.event_name == 'workflow_dispatch' }}
+          PRR_PR_NUMBER: ${{ github.event_name == 'pull_request' && github.event.pull_request.number || steps.resolve-pr.outputs.pr_number }}
+          ELIZACLOUD_API_KEY: ${{ secrets.ELIZACLOUD_API_KEY }}
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+        run: |
+          node dist/tools/prr/index.js \
+            "https://github.com/${{ github.repository }}/pull/${{ github.event_name == 'pull_request' && github.event.pull_request.number || steps.resolve-pr.outputs.pr_number }}" \
+            --tool llm-api \
+            --no-wait-bot
+
+      # output.log + prompts.log are written to CWD (repo root) by PRR; upload as artifacts for debugging.
+      - name: Upload logs
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: prr-logs-${{ github.event_name == 'pull_request' && github.event.pull_request.number || steps.resolve-pr.outputs.pr_number }}
+          path: |
+            output.log
+            prompts.log
+          retention-days: 14
+          if-no-files-found: ignore