← Back to list

pipeline
by oornnery
My custom agent and skills
⭐ 0🍴 0📅 Jan 16, 2026
SKILL.md
name: pipeline description: End-to-end autonomous pipeline from feature request to merged-ready PR. Use when the user says "build/implement/add feature", "fix bug", "create PR", "ship this", or wants a full code change workflow (branch -> analyze -> plan -> execute -> validate -> review -> fix -> tests -> PR).
Pipeline — Feature → PR Autonomous Workflow
You are running a 10-step autonomous engineering pipeline that turns a feature request into a merge-ready PR.
Operating Principles (non-negotiable)
- Small diffs, frequent checkpoints (avoid "big bang" refactors).
- Plan before you write.
- Always validate locally (lint, type-check, tests) before finishing.
- Prefer boring solutions over cleverness.
- Never silently skip failing checks. If something fails: diagnose → fix → re-run.
- Maintain repo conventions (tooling, naming, architecture, patterns).
Modes
| Mode | Description |
|---|---|
AUTO | Fully autonomous; ask only when blocked |
ECONOMY | Minimize context/tool calls; smallest viable change |
INTERACTIVE | Micro-questions at key checkpoints |
If the user did not specify mode, default to AUTO.
Stack Detection
Before planning/executing, infer the stack by scanning:
- lockfiles / package managers (
uv.lock,pnpm-lock.yaml,go.sum,Cargo.lock) - tool configs (
pyproject.toml,package.json,go.mod,Cargo.toml) - CI pipeline files (
.github/workflows/) - existing commands in README/Makefile
Then follow the matching stack guide in stacks/.
Step Execution Order
Execute steps sequentially. Gate steps are blocking.
| Step | File | Purpose | Gate? |
|---|---|---|---|
| 0 | step-00-init.md | Branch + workspace setup | |
| 0b | step-00b-economy.md | Economy mode rules (optional) | |
| 0b | step-00b-interactive.md | Interactive mode rules (optional) | |
| 1 | step-01-analyze.md | Understand system + touch points | |
| 2 | step-02-plan.md | Implementation plan + checkpoints | |
| 3 | step-03-execute.md | Implement in small diffs | |
| 4 | step-04-validate.md | Lint, typecheck, quick tests | |
| 5 | step-05-review.md | Self-PR review | 🚧 GATE |
| 6 | step-06-resolve.md | Fix issues from review | |
| 7 | step-07-tests.md | Add/update tests | |
| 8 | step-08-run-tests.md | Full test suite | 🚧 GATE |
| 9 | step-09-finish.md | PR-ready output |
Gate Steps (Blocking)
These steps require agent delegation and cannot be skipped:
Step 05 — Review Gate
- Delegate to:
revieweragent (READ-ONLY) - Blocking condition: Any
must_fixissue - Required output:
review_passed: truein PIPELINE_STATE
Step 08 — Validation Gate
- Delegate to:
testeragent - Blocking condition: Any test/lint/type failure
- Required output:
validation_passed: truein PIPELINE_STATE
Required Output: PIPELINE_STATE
Every step MUST end with a valid PIPELINE_STATE.
{
"current_step": "step-XX-name",
"goal": "one-line goal",
"mode": "auto|economy|interactive",
"branch": "feat/xxx",
"stack": {
"detected": "python-uv",
"test_command": "uv run pytest -v",
"lint_command": "uv run ruff check .",
"format_command": "uv run ruff format .",
"typecheck_command": "uv run ty check"
},
"files_touched": [],
"test_status": "pending|pass|fail",
"review_passed": true,
"validation_passed": true,
"next_step": "step-XX"
}
See outputs/schema.json for full schema.
Quality Bar to Finish
You can only finish if:
- ✅ Formatting + lint is clean
- ✅ Type-check is clean (if exists)
- ✅ Tests pass (existing + new)
- ✅ Review gate passed (
review_passed: true) - ✅ Validation gate passed (
validation_passed: true) - ✅ PR description complete (what/why/how/tested/risk)
- ✅ No obvious TODOs left behind
Usage Examples
# Full auto (feature → PR)
/pipeline --auto add JWT auth to API
# Quick economy mode
/pipeline --economy fix typo in config loader
# Interactive for risky changes
/pipeline --interactive refactor billing pipeline safely
Subagents (Agent Isolation)
| Agent | Purpose | Tools | Isolation |
|---|---|---|---|
runner | Orchestrates pipeline | Read, Edit, Bash, Grep | Main driver |
reviewer | Code review gate | Read, Grep | READ-ONLY |
tester | Test/validation gate | Read, Edit, Bash | Tests only |
researcher | Codebase analysis | Read, Grep, Glob | READ-ONLY |
Isolation Rules:
- Runner cannot do code review → must delegate to
reviewer - Runner should delegate tests → to
tester - Reviewer cannot edit files
- Researcher cannot edit files
Score
Total Score
50/100
Based on repository quality metrics
✓SKILL.md
SKILL.mdファイルが含まれている
+20
○LICENSE
ライセンスが設定されている
0/10
○説明文
100文字以上の説明がある
0/10
○人気
GitHub Stars 100以上
0/15
○最近の活動
3ヶ月以内に更新がある
0/10
○フォーク
10回以上フォークされている
0/5
✓Issue管理
オープンIssueが50未満
+5
✓言語
プログラミング言語が設定されている
+5
○タグ
1つ以上のタグが設定されている
0/5
Reviews
💬
Reviews coming soon