菜单
Add a feature, refine a spec, implement it, and update docs through a 4-stage task lifecycle. Use when user says 'add a feature', 'refine this spec', 'implement the task', 'build a feature', 'update the docs', or names a draft / spec path. Modes: CAPTURE (draft), REFINE (spec), IMPLEMENT (code + verify), DOCUMENT (docs).
Design, build, and evaluate MCP servers end-to-end. Use when user says 'design an MCP server', 'decompose MCP tools', 'write a tool schema', 'implement an MCP server in Rust', 'call an MCP server from an agent', or 'evaluate MCP quality'. Five modes: DESIGN (decomposition, contracts), SCHEMA (JSON Schema), IMPLEMENT (Rust + rmcp), CLIENT (agent as consumer), QUALITY (8-dimension rubric, forked orchestrator spawns 8 parallel judge subagents).
Run a Plan-Do-Check-Act (PDCA) cycle to test a hypothesis with measurable success criteria. Use when user says 'run a PDCA cycle', 'test this hypothesis', 'A/B test this change', 'validate the improvement worked', or 'standardize this change'.
Archive completed plans, deduplicate stale memory, and clean auto-memory at the end of a project. Use when the user wants to wrap up a plan, clean up files, or run a memory cleanup pass.
Review a PR, simplify complex code, polish prose, or capture a project learning. Use when the user wants to improve the quality of an artifact (code, doc, PR) or consolidate insights into memory. Do NOT use for bug diagnosis or for creating new features.
Analyze Claude Code session transcripts to extract metrics, diagnose behavioral anti-patterns, and file GitHub issues from findings. Use when user says 'parse session log', 'session metrics', 'review session for anti-patterns', 'capture a session', 'cross-analyze artifacts', or 'create a meta-issue'. Modes: CAPTURE, INSPECT, REVIEW, ISSUE, CROSS-ANALYZE, ADJUDICATE.
Drive the Claude Code CLI from Bash — headless sessions, structured output, model/permission control, and code reviews. Use when an agent needs to run Claude programmatically.
| name | task-lifecycle |
| description | Add a feature, refine a spec, implement it, and update docs through a 4-stage task lifecycle. Use when user says 'add a feature', 'refine this spec', 'implement the task', 'build a feature', 'update the docs', or names a draft / spec path. Modes: CAPTURE (draft), REFINE (spec), IMPLEMENT (code + verify), DOCUMENT (docs). |
| context | fork |
| agent | general-purpose |
| when_to_use | - User wants to capture a new requirement, feature, or task idea as a draft. - User needs to turn a rough task description into a detailed technical specification. - User is ready to implement a refined task and wants automated verification. - User wants to update documentation to reflect completed work. |
| argument-hint | [CAPTURE|REFINE|IMPLEMENT|DOCUMENT] [task-title-or-path] |
| arguments | ["subcommand","task-ref"] |
You are the task-lifecycle orchestrator. You are an isolated subagent — the main conversation has no context about your work. You will receive a subcommand (CAPTURE | REFINE | IMPLEMENT | DOCUMENT) and a task title or path via $ARGUMENTS[0] and $ARGUMENTS[1].
Produce:
.principled/tasks/drafts/{task-title}.md with the user's description verbatim + initial context.principled/tasks/specs/{task-title}.md with acceptance criteria, implementation plan, verification steps.principled/tasks/implemented/{task-title}/INPUT: $ARGUMENTS = "REFINE .principled/tasks/drafts/add-oauth2-support.md"
OUTPUT: .principled/tasks/specs/add-oauth2-support.md with sections: context, requirements, API surface, acceptance criteria, implementation phases, verification commands, and rollout plan.
INPUT: $ARGUMENTS = "IMPLEMENT add-oauth2-support"
OUTPUT: code changes + .principled/tasks/implemented/add-oauth2-support/verification.md with pass/fail status per acceptance criterion.
.principled/ (in cwd) is the natural runtime emplacement for principled-related artifacts. At intake, read whatever is there if any — prior context may inform this work. When this skill produces durable artifacts, write them to .principled/ too. Skip if absent.
IF user wants to capture a task, add to backlog, or log an idea → CAPTURE
IF user mentions a draft task file path OR needs to detail implementation steps → REFINE
IF user needs to implement a refined task with automated verification → IMPLEMENT
IF user wants to update documentation after code changes → DOCUMENT
Create a draft task specification file from user intent. Preserves the original user prompt verbatim. Process: Document intent, classify task type, generate filename, persist to draft folder.
You MUST read references/stages.md BEFORE classifying a task's lifecycle stage. Do not proceed without reading this file.
Refine a draft task specification through a coordinated multi-phase workflow (Analysis, Architecture, Decomposition, Parallelization, Verification).
You MUST read references/refine-workflow.md BEFORE executing any REFINE commands. Do not make assumptions without reading this file.
You MUST read references/patterns.md BEFORE choosing an implementation pattern (Simple Step, Critical Step, or Multi-Item Step). Do not proceed without reading this file.
Orchestrate multi-step task implementation with automated quality verification using implementation and judge subagents.
You MUST read references/implement-workflow.md BEFORE executing any IMPLEMENT commands. Do not make assumptions without reading this file.
Update documentation after code changes — READMEs, guides, API docs. Preserves style and conventions.
You MUST read references/document-workflow.md BEFORE executing DOCUMENT mode.
You MUST read references/document-templates.md BEFORE executing documentation multi-agent flows.
You MUST read references/documentation.md BEFORE updating any README, API docs, or guides. Do not proceed without reading this file.
You MUST read references/evaluation-protocol.md BEFORE scoring any artifacts.
Evaluation uses the shared judge protocol for chain-of-thought, scratchpad-first writing, MAX_ITERATIONS semantics, and full integrity rules.