catsofsuffering/ccsm/spec-fast

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Orchestrate the full spec-driven workflow (spec-init to spec-review) automatically with bounded rework loops. Use when Codex should drive the complete change lifecycle without manual phase handoffs.

Install

npx vskill@latest install catsofsuffering/ccsm/spec-fast

SKILL.md

Orchestrate the full spec-driven workflow from Codex with state-aware resume and bounded rework.

Core contract

  • Codex remains the orchestrator across all phases.
  • OpenSpec remains the source of truth for change state and phase boundaries.
  • spec-fast is an orchestration skill; it delegates to existing phase skills (spec-init, spec-plan, spec-impl, spec-review) rather than implementing parallel execution paths.
  • The first implementation attempt is followed by at most 2 rework rounds before the change stops for manual intervention.
  • Codex owns all acceptance and archive decisions; Claude must not decide final acceptance or trigger archive.
  • Archive happens manually after Codex confirms archive-ready, not automatically.

State model

Each change tracks phase progression as a linear sequence:

  1. proposal - Change artifact created
  2. planning - Design and tasks artifacts ready
  3. implementation - Execution in progress or complete
  4. review - Acceptance gate passed or failed

A phase is pending if its artifacts are not yet created, ready if artifacts exist but execution has not started, complete if the phase succeeded, or failed if rework is needed.

Rework model

  • Rework budget: 2 rounds after the initial implementation attempt
  • Attempt 1: Initial spec-impl
  • Attempt 2: First rework after review failure
  • Attempt 3: Second rework after review failure (final attempt)
  • After attempt 3 fails review, the change stops as retry-budget-exhausted

Steps

  1. Confirm OpenSpec is available with openspec --version.
  2. Inspect current changes with openspec list --json.
  3. Select or create the active change:
  • If no suitable change exists, create one with openspec new change "<change-name>".
  • Run openspec status --change "<change-name>" --json to get the current phase state.
  1. Determine the first pending or ready phase by inspecting OpenSpec artifact state.
  2. Resume from that phase by invoking the corresponding skill:
  • spec-init if proposal is missing or stale
  • spec-plan if planning artifacts are not ready
  • spec-impl if implementation is pending or ready
  • spec-review if review is pending or ready
  1. After each phase completes, re-inspect OpenSpec state before proceeding to the next phase.
  2. After spec-impl, run spec-review to verify the implementation.
  3. If review passes: mark the change archive-ready and surface the archive approval step for Codex.
  4. If review fails with a rework packet and rework budget remains:
  • Increment the attempt counter
  • Apply the rework packet to the relevant OpenSpec artifacts
  • Re-run spec-impl with the updated context
  • Re-run spec-review
  1. If review fails and rework budget is exhausted: stop as retry-budget-exhausted and surface the blocker for manual intervention.
  2. If planning, dispatch, or review cannot proceed due to external blockers: stop as blocked and surface the blocker.

Output

After each orchestration cycle, surface:

  • active change id
  • current phase
  • attempt number (if past initial impl)
  • final state (in-progress, archive-ready, blocked, retry-budget-exhausted)
  • next manual step (if any)
  • latest execution packet context (if impl was run)
  • latest review decision with failure reasons (if review failed)

Guardrails

  • Do not run spec-init, spec-plan, spec-impl, or spec-review directly from Claude; always invoke them through Codex orchestration.
  • Do not let Claude decide final acceptance or trigger archive.
  • Do not skip OpenSpec state checks between phases.
  • Do not continue to the next phase without confirming the current phase reached a terminal state in OpenSpec.
  • Do not exceed the rework budget of 2 rounds after initial implementation.
  • Reuse existing spec-impl execution rules and spec-review acceptance criteria verbatim; do not invent parallel acceptance paths.
  • If a phase skill reports blocked, stop immediately and surface the blocker rather than attempting to work around it.
  • Keep all change state mutations in OpenSpec; do not track phase progress outside the OpenSpec artifact model.

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/catsofsuffering/ccsm/spec-fast/badge)](https://verified-skill.com/skills/catsofsuffering/ccsm/spec-fast)

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