agentsclimarketplace

Specpowers flow

Skill ujffdi/specpowers-flow/skills/specpowers-flow

Use when running an end-to-end spec-driven change from idea to archive — orchestrates the full OpenSpec + Superpowers-style workflow with explicit stages and gates. Triggers - "run the full specpowers flow", "start a complete spec-driven change", "go from brainstorm to archive", "use specpowers for this feature".From its SKILL.md

Install
npx -y skills add ujffdi/specpowers-flow --skill specpowers-flow

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

7.5 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it

1. Trigger conditions and new vs resume

Invoke this skill when the user wants to drive a change through the full 8-stage lifecycle. Determine whether this is a new change or a resume:

  • New: no openspec/changes/<change>/ directory exists yet, or the user explicitly requests a fresh start. Proceed from stage 1 (brainstorm).
  • Resume: the change directory exists. Run stage detection (see §2) to identify the current stage and any invalidated gates, then continue from the earliest unresolved stage.

If the change name is ambiguous or multiple active changes exist, ask the user to confirm before proceeding.

2. Resume and stage detection

To determine current stage:

  1. Scan openspec/changes/<change>/ for artifacts: presence of proposal.md, design.md, tasks.md, specs/, and coverage/compliance markers.
  2. Scan .specpowers/gates/ inside the change directory for per-stage gate evidence files (see references/stage-protocol.md for the evidence record shape: stage, passed_at, artifacts[{path, sha256}], result).
  3. Recompute content digests for every artifact referenced in each gate evidence file. If a digest no longer matches disk, invalidate that gate and all downstream gates and route back to the earliest invalidated stage.
  4. .specpowers-state.yaml is a hint cache only — never treat it as authoritative. Always derive stage from on-disk artifacts and gate evidence.

Full stage definitions, evidence record format, and invalidation rules are in references/stage-protocol.md. Read that file when resuming a change or debugging a stale gate.

3. Tier selection

Before advancing past brainstorm, select a tier — quick, standard, or full — using the rules in references/tiering-rules.md. The orchestrator estimates size (files likely touched, reversibility, blast radius) and suggests a default; the user may override downward only within the limits defined there.

Non-overridable escalation: any change touching authentication/authorization/permissions, data migration or schema changes, destructive or irreversible state, tenant/security boundaries, or money/billing is forced to standard or full regardless of size or user preference. This cannot be bypassed. Such changes also require a real spec delta before coverage/compliance can pass.

Behavioral-change delta rule: any change that alters system behavior requires a real spec delta in every tier, including quick. Without a delta the coverage and compliance gates have no contract to verify against. The no-spec-delta exception applies only to genuinely non-behavioral changes (docs/formatting/comments). Read references/tiering-rules.md at tier selection time and again when coverage or compliance gates are evaluated.

4. The 8-stage state machine

Each stage has a mandatory completion gate. No stage may be marked complete until its gate passes. The phase skill responsible for each stage is listed below; route to it and hand back control to the orchestrator when it completes or fails.

StageGate (summary)Phase skill
1. brainstormDirection approved; requirement specific enough to generate a changespecpowers-brainstorm
2. generate-specChange dir exists; proposal.md, design.md, tasks.md, and spec deltas presentspecpowers-spec
3. harden-specValidation passes; no unresolved blocker; findings synced back into artifactsspecpowers-spec
4. plan-from-specPlan exists in tasks.md; explicitly based on the hardened specspecpowers-plan
5. check-coverageEvery requirement has ≥1 plan step and ≥1 verification path; coverage matrix completespecpowers-plan
6. execute-planImplementation complete; tests run (test-first sub-gate satisfied); evidence preservedspecpowers-build
7. verify-complianceCompliance passes; tests pass; no unresolved blockerspecpowers-build
8. archiveAll 7 prior gates passed (evidence digests still match disk); user confirmation when required by tierspecpowers-archive

Full input/output/failure-handling details for every stage are in references/stage-protocol.md.

5. Gate enforcement and failure routing

Never advance past a stage whose gate has not passed. When a gate fails, route back to the stage indicated by the failure-routing table in references/stage-protocol.md. Common cases:

  • Validation or artifact check fails → route to generate-spec
  • Adversarial spec review finds blockers → sync findings, re-run harden-spec
  • Coverage gap → route to plan-from-spec
  • Task implementation diverges from spec → update artifact first (which invalidates downstream gates), then resume from the updated stage
  • Tests fail or compliance fails → route to execute-plan or verify-compliance as appropriate
  • Archive requested before prior gates pass → block; route back to the first failing gate

Do not proceed to the next stage until the user has been informed of the failure and the affected stage is resolved.

6. When to read each reference file

ReferenceRead when
references/stage-protocol.mdOn every invoke: stage detection, gate evaluation, failure routing, evidence record format
references/tiering-rules.mdAt tier selection; again when evaluating coverage or compliance gates
references/openspec-artifact-format.mdWhen creating or validating the change directory and its artifacts
references/adversarial-spec-review.mdWhen running the harden-spec adversarial review
references/independent-review.mdWhen dispatching any adversarial subagent (harden-spec, verify-compliance)
references/subagent-execution.mdBefore executing the plan in execute-plan
references/test-driven-development.mdBefore each execute-plan task and when evaluating the test-first sub-gate
references/plan-coverage-matrix.mdAt check-coverage and when verifying the compliance inputs
references/compliance-verification.mdAt verify-compliance stage
references/archive-checklist.mdAt archive stage; includes conservative fallback archive procedure

7. Progressive enhancement

At startup, probe for the presence of real tools:

  • openspec CLI present (openspec --version succeeds): use openspec validate for artifact validation and openspec archive for the archive merge. The format contract remains references/openspec-artifact-format.md.
  • Superpowers skills present: when executing the plan, hand off to Superpowers subagent-driven-development or executing-plans rather than the self-contained protocol in references/subagent-execution.md. Similarly, hand off brainstorm to Superpowers brainstorming and test-first execution to Superpowers test-driven-development when present.
  • Neither present: use the built-in markdown procedures defined in the reference files. This is the default fallback path and is fully functional.

The skill runs correctly in all three configurations. The self-contained fallback paths in references/subagent-execution.md, references/test-driven-development.md, and references/archive-checklist.md are authoritative when real tools are absent.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,750. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.