CodingThrust · GiggleLiu · Mar 12, 2026 · Mar 11, 2026 · Mar 11, 2026 · Mar 12, 2026
diff --git a/.claude/CLAUDE.md b/.claude/CLAUDE.md
@@ -15,6 +15,8 @@ Rust library for NP-hard problem reductions. Implements computational problems w
 - [check-issue](skills/check-issue/SKILL.md) -- Quality gate for `[Rule]` and `[Model]` issues. Checks usefulness, non-triviality, correctness of literature, and writing quality. Posts structured report and adds failure labels.
 - [check-rule-redundancy](skills/check-rule-redundancy/SKILL.md) -- Check if a reduction rule (source-target pair) is redundant, i.e., dominated by a composite path through other rules.
 - [meta-power](skills/meta-power/SKILL.md) -- Batch-resolve all open `[Model]` and `[Rule]` issues autonomously: plan, implement, review, fix CI, merge — in dependency order (models first).
+- [project-pipeline](skills/project-pipeline/SKILL.md) -- Pick a Ready issue from the GitHub Project board, move it through In Progress -> issue-to-pr --execute -> review-agentic.
+- [review-pipeline](skills/review-pipeline/SKILL.md) -- Pick a PR from review-agentic column, fix Copilot review comments, fix CI, run agentic feature tests, move to In Review.
 
 ## Commands
 ```bash
@@ -42,6 +44,10 @@ make cli-demo      # Run closed-loop CLI demo (exercises all commands)
 make mcp-test      # Run MCP server tests (unit + integration)
 make run-plan      # Execute a plan with Claude autorun
 make run-issue N=42 # Run issue-to-pr --execute for a GitHub issue
+make run-pipeline  # Pick next Ready issue from project board, implement, move to review-agentic
+make run-pipeline N=97 # Process a specific issue from the project board
+make run-review    # Pick next PR from review-agentic column, fix Copilot comments, fix CI, run agentic tests
+make run-review N=570 # Process a specific PR from the review-agentic column
 make copilot-review # Request Copilot code review on current PR
 make release V=x.y.z  # Tag and push a new release (CI publishes to crates.io)
 ```

diff --git a/.claude/skills/issue-to-pr/SKILL.md b/.claude/skills/issue-to-pr/SKILL.md
@@ -203,62 +203,13 @@ git push
 make copilot-review
 ```
 
-#### 7e. Fix Loop (max 3 retries)
-
-```bash
-REPO=$(gh repo view --json nameWithOwner --jq .nameWithOwner)
-```
-
-For each retry:
-
-1. **Wait for CI to complete** (poll every 30s, up to 15 minutes):
-   ```bash
-   for i in $(seq 1 30); do
-       sleep 30
-       HEAD_SHA=$(gh api repos/$REPO/pulls/$PR | python3 -c "import sys,json; print(json.load(sys.stdin)['head']['sha'])")
-       STATUS=$(gh api repos/$REPO/commits/$HEAD_SHA/check-runs | python3 -c "
-   import sys,json
-   runs = json.load(sys.stdin)['check_runs']
-   if not runs:
-       print('PENDING')  # CI hasn't registered yet
-   else:
-       failed = [r['name'] for r in runs if r.get('conclusion') not in ('success', 'skipped', None)]
-       pending = [r['name'] for r in runs if r.get('conclusion') is None and r['status'] != 'completed']
-       if pending:
-           print('PENDING')
-       elif failed:
-           print('FAILED')
-       else:
-           print('GREEN')
-   ")
-       if [ "$STATUS" != "PENDING" ]; then break; fi
-   done
-   ```
-
-   - If `GREEN` on the **first** iteration (before any fix-pr): skip the fix loop, done.
-   - If `GREEN` after a fix-pr pass: break, done.
-   - If `FAILED`: continue to step 2.
-   - If still `PENDING` after 15 min: treat as `FAILED`.
-
-2. **Invoke `/fix-pr`** to address review comments, CI failures, and coverage gaps.
-
-3. **Push fixes:**
-   ```bash
-   git push
-   ```
-
-4. Increment retry counter. If `< 3`, go back to step 1. If `= 3`, give up.
-
-**After 3 failed retries:** leave PR open, report to user.
-
-#### 7f. Done
+#### 7e. Done
 
 Report final status:
-- PR URL
-- CI status (green / failed after retries)
-- Any unresolved review items
+- PR URL and number
+- Implementation summary
 
-The PR is **not merged** — the user or `meta-power` decides when to merge.
+The PR is **not merged** and CI/review fixes are **not** handled here. The separate `review-pipeline` skill picks up PRs from the `review-agentic` board column to handle Copilot review comments, CI fixes, and agentic testing.
 
 ## Example
 
@@ -286,9 +237,9 @@ Executing plan via subagent-driven-development...
 [Subagents implement the plan steps]
 [Runs review-implementation — all checks pass, auto-fixes applied]
 [Pushes + requests Copilot review]
-[Polls CI... GREEN on first pass]
 
-PR #45: CI green, ready for merge.
+PR #45 created and pushed. Copilot review requested.
+Run /review-pipeline to process Copilot comments, fix CI, and run agentic tests.
 ```
 
 ## Common Mistakes

diff --git a/.claude/skills/project-pipeline/SKILL.md b/.claude/skills/project-pipeline/SKILL.md
@@ -0,0 +1,175 @@
+---
+name: project-pipeline
+description: Pick a Ready issue from the GitHub Project board, move it through In Progress -> issue-to-pr -> review-agentic
+---
+
+# Project Pipeline
+
+Pick a "Ready" issue from the [GitHub Project board](https://github.com/orgs/CodingThrust/projects/8/views/1), move it to "In Progress", run `issue-to-pr --execute`, then move it to "review-agentic". The separate `review-pipeline` handles Copilot comments, CI fixes, and agentic testing.
+
+## Invocation
+
+- `/project-pipeline` -- pick the next Ready issue (first Model, then Rule, by issue number)
+- `/project-pipeline 97` -- process a specific issue number from the Ready column
+- `/project-pipeline --all` -- batch-process all Ready issues (Models first, then Rules)
+
+## Constants
+
+GitHub Project board IDs (for `gh project item-edit`):
+
+| Constant | Value |
+|----------|-------|
+| `PROJECT_ID` | `PVT_kwDOBrtarc4BRNVy` |
+| `STATUS_FIELD_ID` | `PVTSSF_lADOBrtarc4BRNVyzg_GmQc` |
+| `STATUS_READY` | `61e4505c` |
+| `STATUS_IN_PROGRESS` | `47fc9ee4` |
+| `STATUS_REVIEW_AGENTIC` | `b2f16561` |
+| `STATUS_IN_REVIEW` | `df73e18b` |
+| `STATUS_DONE` | `98236657` |
+
+## Autonomous Mode
+
+This skill runs **fully autonomously** — no confirmation prompts, no user questions. It picks the next issue and processes it end-to-end. All sub-skills (`issue-to-pr`, `check-issue`, `add-model`, `add-rule`, etc.) should also auto-approve any confirmation prompts.
+
+## Steps
+
+### 0. Discover Ready Issues
+
+```bash
+gh project item-list 8 --owner CodingThrust --format json
+```
+
+Filter items where `status == "Ready"`. Partition into `[Model]` and `[Rule]` buckets, sort each by issue number ascending. Final order: **all Models first, then all Rules** (so dependencies are satisfied).
+
+Print the list for visibility (no confirmation needed):
+
+```
+Ready issues:
+  Models:
+    #129  [Model] MultivariateQuadratic
+    #117  [Model] GraphPartitioning
+  Rules:
+    #97   [Rule] BinPacking to ILP
+    #110  [Rule] LCS to ILP
+    #126  [Rule] KSatisfiability to SubsetSum
+    #130  [Rule] MultivariateQuadratic to ILP
+```
+
+**If a specific issue number was provided:** verify it is in the Ready column. If not, STOP with a message.
+
+**If `--all`:** proceed immediately with all Ready issues in order (no confirmation).
+
+**Otherwise (no args):** pick the first issue in the ordered list (Models before Rules, lowest number first) and proceed immediately (no confirmation).
+
+### 1. Create Worktree
+
+Create an isolated git worktree for this issue so the main working directory stays clean:
+
+```bash
+git fetch origin main
+BRANCH="issue-<number>-<slug>"
+WORKTREE_DIR=".worktrees/$BRANCH"
+mkdir -p .worktrees
+git worktree add "$WORKTREE_DIR" -b "$BRANCH" origin/main
+cd "$WORKTREE_DIR"
+```
+
+All subsequent steps run inside the worktree. This ensures the user's main checkout is never modified.
+
+### 2. Move to "In Progress"
+
+Extract the project item ID for the chosen issue from the JSON output (the `id` field of the matching item).
+
+```bash
+gh project item-edit \
+  --id <ITEM_ID> \
+  --project-id PVT_kwDOBrtarc4BRNVy \
+  --field-id PVTSSF_lADOBrtarc4BRNVyzg_GmQc \
+  --single-select-option-id 47fc9ee4
+```
+
+### 3. Run issue-to-pr --execute
+
+Invoke the `issue-to-pr` skill with `--execute` (working directory is the worktree):
+
+```
+/issue-to-pr <number> --execute
+```
+
+This handles the full pipeline: fetch issue, verify Good label, research, write plan, create PR, implement, review, fix CI.
+
+**If `issue-to-pr` fails:** record the failure, but still move the issue to "In Review" so it's visible for human triage. Report the failure to the user.
+
+### 4. Move to "review-agentic"
+
+After `issue-to-pr` completes (success or failure with a PR created), move the issue to the `review-agentic` column for the second-stage review pipeline:
+
+```bash
+gh project item-edit \
+  --id <ITEM_ID> \
+  --project-id PVT_kwDOBrtarc4BRNVy \
+  --field-id PVTSSF_lADOBrtarc4BRNVyzg_GmQc \
+  --single-select-option-id b2f16561
+```
+
+**If no PR was created** (issue-to-pr failed before creating a PR): move the issue back to "Ready" instead:
+
+```bash
+gh project item-edit \
+  --id <ITEM_ID> \
+  --project-id PVT_kwDOBrtarc4BRNVy \
+  --field-id PVTSSF_lADOBrtarc4BRNVyzg_GmQc \
+  --single-select-option-id 61e4505c
+```
+
+### 5. Clean Up Worktree
+
+After the issue is processed (success or failure), clean up the worktree:
+
+```bash
+cd /Users/liujinguo/rcode/problemreductions
+git worktree remove "$WORKTREE_DIR" --force
+```
+
+### 6. Report (single issue)
+
+Print a summary:
+
+```
+Pipeline complete:
+  Issue:  #97 [Rule] BinPacking to ILP
+  PR:     #200
+  Status: Awaiting agentic review
+  Board:  Moved Ready -> In Progress -> review-agentic
+```
+
+### 7. Batch Mode (`--all`)
+
+If `--all` was specified, repeat Steps 1-6 for each issue in order. Each issue gets its own worktree (created and cleaned up per issue).
+
+After all issues, print a batch report:
+
+```
+=== Project Pipeline Batch Report ===
+
+| Issue | Title                              | PR   | Status      | Board       |
+|-------|------------------------------------|------|-------------|-------------|
+| #129  | [Model] MultivariateQuadratic      | #201 | CI green    | review-agentic |
+| #97   | [Rule] BinPacking to ILP           | #202 | CI green    | review-agentic |
+| #110  | [Rule] LCS to ILP                  | #203 | fix failed  | review-agentic |
+| #126  | [Rule] KSat to SubsetSum           | -    | plan failed | Ready       |
+
+Completed: 2/4 | In Review: 3 | Returned to Ready: 1
+```
+
+## Common Mistakes
+
+| Mistake | Fix |
+|---------|-----|
+| Issue not in Ready column | Verify status before processing; STOP if not Ready |
+| Missing project scopes | Run `gh auth refresh -s read:project,project` |
+| Forgetting to move back to Ready on total failure | Only move to In Review if a PR exists |
+| Processing Rules before Models | Always sort Models first — Rules may depend on them |
+| Not syncing main between batch issues | Each issue gets a fresh worktree from `origin/main` |
+| Worktree left behind on failure | Always clean up with `git worktree remove` in Step 5 |
+| Working in main checkout | All work happens in `.worktrees/` — never modify the main checkout |