Back to list
oornnery

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

ModeDescription
AUTOFully autonomous; ask only when blocked
ECONOMYMinimize context/tool calls; smallest viable change
INTERACTIVEMicro-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.

StepFilePurposeGate?
0step-00-init.mdBranch + workspace setup
0bstep-00b-economy.mdEconomy mode rules (optional)
0bstep-00b-interactive.mdInteractive mode rules (optional)
1step-01-analyze.mdUnderstand system + touch points
2step-02-plan.mdImplementation plan + checkpoints
3step-03-execute.mdImplement in small diffs
4step-04-validate.mdLint, typecheck, quick tests
5step-05-review.mdSelf-PR review🚧 GATE
6step-06-resolve.mdFix issues from review
7step-07-tests.mdAdd/update tests
8step-08-run-tests.mdFull test suite🚧 GATE
9step-09-finish.mdPR-ready output

Gate Steps (Blocking)

These steps require agent delegation and cannot be skipped:

Step 05 — Review Gate

  • Delegate to: reviewer agent (READ-ONLY)
  • Blocking condition: Any must_fix issue
  • Required output: review_passed: true in PIPELINE_STATE

Step 08 — Validation Gate

  • Delegate to: tester agent
  • Blocking condition: Any test/lint/type failure
  • Required output: validation_passed: true in 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)

AgentPurposeToolsIsolation
runnerOrchestrates pipelineRead, Edit, Bash, GrepMain driver
reviewerCode review gateRead, GrepREAD-ONLY
testerTest/validation gateRead, Edit, BashTests only
researcherCodebase analysisRead, Grep, GlobREAD-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