claude-skills

Execute a task WITH tests (the test-inclusive version of code-execute). Pipeline: codex implements → test-plan-agent corrects test-plan.md → codex (gpt-5.4) writes tests → codex runs tests → triple-review. Invoke this skill — not code-execute — any time the user wants to execute/run a task and tests are in scope. This includes when the user mentions TDD, 测试策略, writing tests, test strategy, or their Obsidian task folder has a test-plan.md. Common phrasings: '执行+测试', '走 TDD 流程', 'TDD 流水线', '需要写测试', 'execute with TDD'. Needs an existing Obsidian plan + test-plan.md (produced by obsidian-plan-tdd). Does not plan tests or commit code.

Execute a task by handing coding to codex:execute and review to triple-review. Three modes: (1) tracker task-id with Obsidian plan, (2) Obsidian lookup by folder path or title query, (3) pure prompt. Does NOT commit.

Walk a Jira issue forward along the project's dev workflow until it reaches Dev Complete. Designed for SPFC project's multi-step transition chain (Backlog → 待办 → 正在进行 → 准备审查 → Dev Complete). Reads the current status, then applies the hardcoded happy-path transitions one by one. Stops and asks if the current state is outside the known chain. Does NOT take the issue beyond Dev Complete.

Execute a Jira task by handing the coding work to codex:rescue and the review work to the claude-code-review subagent. Pulls issue context via jira-cli, verifies the vault has design + execution plan docs, checks the working tree is clean, dispatches codex:rescue as the implementer, then launches claude-code-review (Opus) to verify the diff against the plan. Does NOT commit; commit is a separate action.

Create task planning in the Obsidian vault. Collects task parameters, creates the vault folder structure and skeleton files from templates, then hands off to the plan skill. Supports both tracker-backed tasks (with task-id) and ad-hoc tasks (no task-id, slug derived from title).

Strict code review against the task's design and execution plan in the Obsidian vault. Use after any implementer (codex, human, another agent) has finished a coding round and you need an opinionated, evidence-based review of whether the changes match the plan, follow the target language's best practices, and carry no real bugs / leaks / regressions. Reports findings only — never modifies files, never applies fixes, never decides what gets fixed.

Subagent that writes test code based on test-plan.md. Reads the finalized test strategy, writes tests per the four-layer rules (unit, contract, behavior-level), compile-verifies only. Does not design test strategy, run tests, or fix implementation. Spawned by code-execute-tdd in CT2, not invoked directly by users.

Subagent that designs test strategy and produces test-plan.md. Two modes: (1) initial — reads plan.md and produces test-plan.md based on design intent; (2) correction — reads existing test-plan.md + actual diff, corrects test-plan.md in place. Spawned by vibe-plan-tdd (mode 1) and code-execute-tdd CT2 (mode 2), not invoked directly by users.

Internal guidance for composing prompts that a main agent hands off to Claude Opus subagents (e.g. claude-code-review). Based on Anthropic's official Claude 4.6 prompting best practices. Use when spawning a Claude Opus subagent and you need to structure the forwarded prompt for maximum clarity, groundedness, and reliable structured output.

Translates Figma designs into production-ready application code with 1:1 visual fidelity. Use when implementing UI code from Figma files, when user mentions "implement design", "generate code", "implement component", or asks to build components matching Figma specs. Not for data collection or planning — if user wants to save design data, extract tokens, or plan before coding, use figma-plan instead. For Figma canvas writes via `use_figma`, use `figma-use`.

Post a tester-facing comment on a YouTrack issue via the pp-youtrack MCP server. Reads the task's Obsidian plan docs (root-cause.md + fix-plan.md for bugs, plan.md for features) and summarizes them in a voice appropriate for QA: plain language, just enough technical detail (URLs, modules, scenarios) to be useful, but no deep internals. Comment is always written in Chinese. Narrow scope — only comment posting. State transitions live in pp-youtrack-state.

Transition a YouTrack issue's workflow state (e.g. 进行中, 自测, 完成) through the pp-youtrack MCP server. Narrow-scope skill focused on state transitions only — no comments, no creation, no search. Other YouTrack operations live in their own pp-youtrack-* skills (e.g. pp-youtrack-comment). Used by obsidian-design and code-execute to auto-advance tasks along the lifecycle.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

1 hook across 1 event