skills/ahgraber/skills/spec-kit-specify

spec-kit-specify

SKILL.md

Spec Kit Specify

Create or refresh the feature specification for the active Spec Kit feature.

When to Use

  • Start a new feature and create the first spec.md.
  • Convert free-form requirements into a structured, testable, implementation-agnostic specification.
  • Rework an existing spec draft so it is ready for spec-kit-clarify or spec-kit-plan.

When Not to Use

  • Clarify ambiguity in an already-written spec without redrafting it (spec-kit-clarify).
  • Generate design artifacts or tasks from an approved spec (spec-kit-plan, spec-kit-tasks).
  • Reconcile cross-artifact drift in an existing feature (spec-kit-reconcile).

Router Fit

  • Primary route from spec-kit when spec.md does not exist yet.
  • Downstream: hand off to spec-kit-clarify (if high-impact ambiguity remains) or spec-kit-plan (if spec is ready).

Preconditions

  • User provided a non-empty feature description.
  • Run from repository root (or a subdirectory inside the repository).

Workflow

  1. Normalize input:

    • Treat the full user request as the feature description.
    • If empty, stop: ERROR: No feature description provided.
  2. Generate a short branch suffix (2-4 words, kebab-case):

    • Preserve meaningful technical terms/acronyms.
    • Prefer action-noun phrasing (for example user-auth, bulk-export-audit-log).
  3. Bootstrap feature branch and spec exactly once:

    • Run scripts/create-new-feature.sh --json --short-name "<short-name>" "<feature description>".
    • Do not manually pre-compute branch numbers; let the script assign the next number.
    • Parse JSON output and capture:
      • BRANCH_NAME
      • SPEC_FILE
      • FEATURE_NUM
    • Derive FEATURE_DIR as dirname(SPEC_FILE).
  4. Load template context:

    • Preferred template: {REPO_ROOT}/templates/spec-template.md.
    • Fallback template: assets/spec-template.md.
    • Preserve heading order from the selected template.
  5. Draft spec.md content (focus on WHAT/WHY, not HOW):

    • Fill prioritized user stories with independent tests and acceptance scenarios.
    • Write testable functional requirements.
    • Add edge cases and scope boundaries.
    • Add measurable, technology-agnostic success criteria.
    • Include key entities when data is central.
    • Use reasonable defaults and document assumptions when needed.
  6. Clarification policy while drafting:

    • First, make informed defaults using domain norms.
    • Add [NEEDS CLARIFICATION: ...] only for high-impact uncertainty.
    • Hard limit: at most 3 clarification markers.
    • Prioritize by impact: scope > security/privacy > UX > technical detail.
  7. Write the spec to SPEC_FILE.

  8. Create and run requirements quality validation:

    • Create FEATURE_DIR/checklists/requirements.md.
    • Validate spec against this checklist:
      • No implementation details (frameworks, languages, APIs, internal architecture).
      • Mandatory sections are complete.
      • Requirements are unambiguous and testable.
      • Success criteria are measurable and technology-agnostic.
      • User scenarios and edge cases are covered.
      • Scope boundaries, dependencies, and assumptions are explicit.
      • No unresolved [NEEDS CLARIFICATION] markers for plan-ready specs.
    • If non-clarification issues fail, revise spec and re-validate (max 3 passes).
  9. If clarification markers remain after validation:

    • Ask up to 3 numbered clarification questions total.
    • For each question, present 2-3 concrete options plus a short custom answer path.
    • Wait for user responses, update spec.md, then re-run validation.
  10. Report completion:

  • Branch name.
  • spec.md path.
  • requirements.md path and pass/fail summary.
  • Recommended next step:
    • spec-kit-clarify if high-impact ambiguity remains.
    • spec-kit-plan if ready.

Output

  • Active feature branch from scripts/create-new-feature.sh
  • specs/<feature>/spec.md
  • specs/<feature>/checklists/requirements.md
  • readiness handoff for spec-kit-clarify or spec-kit-plan

Common Mistakes

  • Writing implementation design into the spec instead of user-visible behavior and outcomes.
  • Leaving vague requirements that cannot be acceptance-tested.
  • Asking too many clarification questions instead of making reasonable defaults.
  • Running feature bootstrap script multiple times for one request.

References

  • references/spec-kit-workflow.dot
  • scripts/create-new-feature.sh
  • assets/spec-template.md
  • https://github.com/github/spec-kit/blob/9111699cd27879e3e6301651a03e502ecb6dd65d/templates/commands/specify.md
Weekly Installs
7
Repository
ahgraber/skills
First Seen
Feb 8, 2026
Installed on
opencode7
gemini-cli6
github-copilot6
codex6
continue6
claude-code5