luanmorenommaciel/task-spec/claude-code
Claude Code and agent skill by luanmorenommaciel/task-spec
- community
Checking that the SKILL.md file is still in the repository.
Turn intent and repository evidence into native reviewed plans and atomic, signed Task-Specs; run authorized recipes with TaskMesh, inspect canonical acceptance, and retain release and maintenance evidence.
Install
npx vskill@latest install luanmorenommaciel/task-spec/claude-codeSKILL.md
Task-Spec for Claude Code
Task-Spec is the atomic contract; Claude Code is a replaceable authoring or execution harness. The canonical doctrine ships with the installed engine, and the taskspec CLI owns deterministic validation and mutation.
Read current taskspec agent-context and task or initiative status before acting. Use the same product skill and persisted state as other supported harnesses; conversation history does not authorize work. Detailed workflows are available through taskspec guide intent, decomposition, recipes, execution, acceptance, sdlc, and migration.
Prerequisites
The taskspec CLI must be on PATH and taskspec doctor must pass. Use the repository installer with --global for user-level Claude, Codex/Kimi, and Grok skills, or --target <repo> for project-local copies.
What to do
- Inspect repository instructions, the smallest relevant code slice, its tests,
schemas, and ownership boundaries.
- Use optional research only when explicitly selected; normalize retained
results as credential-free AuthoringEvidence/v1.
- Compose every declared unit in
TaskPlan/v1, preview it, and generate only
the approved plan. Never invent omitted paths, evals, dependencies, or children.
- Drive the CLI; never hand-edit authorization or acceptance fields:
| Intent | Command | |
|---|---|---|
| Prepare native decomposition | taskspec decompose prepare <initiative> --recipe <file> | |
| Explicit topology review | taskspec decompose review <initiative> --accept --reviewer <name> --reason <text> | |
| Compile reviewed plan | taskspec decompose compile <initiative> | |
| Compose and preview | taskspec plan --manifest <TaskPlan> | |
| Generate approved units | taskspec batch --plan <TaskPlan> | |
| Direct scaffold | taskspec new <slug> <effort> [agent] | |
| Prove authoring | taskspec validate <spec> then taskspec dod <spec> | |
| PRE-gate authorization | taskspec gate --stamp <spec> | |
| Portable handoff | taskspec handoff <spec> --backend claude --out <file> | |
| POST-gate acceptance | taskspec accept --handoff <file> --stamp <spec> | |
| Status and recovery | taskspec status <id> and taskspec doctor --backlog | |
| Optional TaskMesh cockpit | taskspec mesh frontier, run, watch, accept, and finish | |
| Managed recipe or native graph | taskspec mesh run --initiative <id> --execute | |
| Opt-in evidence task | taskspec new --format 4 <slug> <effort> then taskspec author-doctor <spec> |
- Respect the effort gate: XS/S/M/L are runnable leaves; L needs a backend
from TASKSPEC_LONG_HORIZON_BACKENDS. XL/XXL are non-runnable nodes with at least two/three child Task-Specs.
Hard rules
signed_off: truecomes ONLY fromtaskspec gate --stamp. Never hand-stamp
the envelope; HMAC v3 seals TaskRevision/v1 and unknown authority fields.
- A task is DONE only when
taskspec acceptemitsACCEPTED=1— an agent's
own claim of GREEN is not evidence. Acceptance must bind TaskHandoff/v3, the attempt, base commit, closure, repository scope, and required receipts.
- Format v4 acceptance must receive every policy-required external receipt;
never expose a private holdout bundle to the executor.
- Changed scope or proof policy requires explicit replanning and reauthorization;
never silently rewrite downstream dependencies.
- Resolve recipe guidance before sealing. Managed recipes use bounded sequential
rounds in the existing TaskMesh lifecycle, including one-task graphs. Workers cannot sign or accept their own work. Reuse applicable recorded decisions and ask only for missing decisions or authorization.
- An incident becomes candidate work. Pipeline success is reported evidence;
it does not establish target health or authorize deployment.
- Exit codes:
0pass/accept ·1failed/rejected ·2usage error. - When TaskMesh is installed, Claude may start or reconnect to the durable
cockpit, but the daemon owns the run and Task-Spec owns authority. Do not call a supervised worktree a sandbox, self-accept an attempt, or merge the target branch. Autonomous OMP requires externally verified isolation and expiring credential evidence.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingharness/claude-code/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/luanmorenommaciel/task-spec/claude-code)