luanmorenommaciel/task-spec/claude-code

Claude Code and agent skill by luanmorenommaciel/task-spec

Trust tier 3
  • community
Source check pending

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-code

SKILL.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

  1. Inspect repository instructions, the smallest relevant code slice, its tests,

schemas, and ownership boundaries.

  1. Use optional research only when explicitly selected; normalize retained

results as credential-free AuthoringEvidence/v1.

  1. Compose every declared unit in TaskPlan/v1, preview it, and generate only

the approved plan. Never invent omitted paths, evals, dependencies, or children.

  1. Drive the CLI; never hand-edit authorization or acceptance fields:
IntentCommand
Prepare native decompositiontaskspec decompose prepare <initiative> --recipe <file>
Explicit topology reviewtaskspec decompose review <initiative> --accept --reviewer <name> --reason <text>
Compile reviewed plantaskspec decompose compile <initiative>
Compose and previewtaskspec plan --manifest <TaskPlan>
Generate approved unitstaskspec batch --plan <TaskPlan>
Direct scaffoldtaskspec new <slug> <effort> [agent]
Prove authoringtaskspec validate <spec> then taskspec dod <spec>
PRE-gate authorizationtaskspec gate --stamp <spec>
Portable handofftaskspec handoff <spec> --backend claude --out <file>
POST-gate acceptancetaskspec accept --handoff <file> --stamp <spec>
Status and recoverytaskspec status <id> and taskspec doctor --backlog
Optional TaskMesh cockpittaskspec mesh frontier, run, watch, accept, and finish
Managed recipe or native graphtaskspec mesh run --initiative <id> --execute
Opt-in evidence tasktaskspec new --format 4 <slug> <effort> then taskspec author-doctor <spec>
  1. 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: true comes ONLY from taskspec gate --stamp. Never hand-stamp

the envelope; HMAC v3 seals TaskRevision/v1 and unknown authority fields.

  • A task is DONE only when taskspec accept emits ACCEPTED=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: 0 pass/accept · 1 failed/rejected · 2 usage 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

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
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/luanmorenommaciel/task-spec/claude-code/badge)](https://verified-skill.com/skills/luanmorenommaciel/task-spec/claude-code)

Something wrong with this skill or its page? Tell us.