AGENT SKILL

speckit-bug-fix

Apply the remediation from a bug assessment and record what was changed

License 许可
MIT
最近更新
2026-11-14
Stars
★ 0
Forks
0

01 前言 Front Matter

name
speckit-bug-fix
metadata
{ "author": "github-spec-kit", "source": "bug:commands/speckit.bug.fix.md" }
description
Apply the remediation from a bug assessment and record what was changed
compatibility
Requires spec-kit project structure with .specify/ directory

03 SKILL.md 正文

Fix Bug

Apply the remediation that was proposed by /speckit-bug-assess and record the changes in a fix report at .specify/bugs/<slug>/fix.md. This command is only valid after an assessment exists for the given slug. Pass --branch (or --worktree) to isolate the fix on its own git branch before editing, mirroring how /speckit-specify isolates feature work.

User Input

$ARGUMENTS

The user input should identify the bug to fix. Accept any of:

  • slug=<bug-slug> or --slug <bug-slug> or just a bare slug-like token.
  • A path that contains the slug (e.g. .specify/bugs/login-timeout/).
  • Branch isolation (optional): branch / --branch creates a fix branch (<prefix>/<slug>) first; worktree / --worktree creates a git worktree instead. See Optional — isolate the fix on a branch below.
  • Nothing — fall back to context (see below).

Slug Resolution

Resolve BUG_SLUG in this order, stopping at the first match:

  1. Explicit user input — a slug passed in $ARGUMENTS (any of the forms above).
  2. Conversation context — if the current session has just run /speckit-bug-assess, the slug it reported is the working slug. Reuse it without re-prompting. Confirm it by checking that .specify/bugs/<slug>/assessment.md exists; if it does not, fall through.
  3. Single candidate on disk — list .specify/bugs/*/assessment.md. If exactly one matching assessment.md is found, use the slug from its parent directory.
  4. Disambiguate:
    • Interactive mode: ask the user which bug to fix and list the candidates.
    • Automated mode: stop with an error listing the candidates. Do not guess.

Once resolved, set BUG_SLUG and BUG_DIR = .specify/bugs/<BUG_SLUG>, and briefly state in your reply which resolution path was used (explicit / from context / single candidate / asked).

Prerequisites

  • BUG_DIR/assessment.md MUST exist. If it does not, stop and instruct the user to run /speckit-bug-assess first.
  • If BUG_DIR/fix.md already exists, ask the user whether to overwrite it before continuing (interactive mode) or refuse (automated mode).
  • Read BUG_DIR/assessment.md in full. Treat its Proposed Remediation, Files likely to change, Tests to add or update, and Risks & Considerations sections as the contract for this command.

TDD mode (when tdd_enabled: true in bug-config.yml)

When the bug extension's tdd_enabled config is true (the default), fix the bug through the TDD extension's red-green-refactor loop instead of editing code directly. This mirrors how spec-whole drives a feature — the "feature" here is the bug:

  1. Ensure the TDD stack profile exists — if .specify/memory/tdd-profile.md is missing, run /speckit.tdd.setup first (every other TDD command requires it).
  2. Synthesize the bug spec — write BUG_DIR/spec.md from assessment.md: the required (fixed) behavior becomes the acceptance criteria, and the assessment's reproduction steps become the failing-test scenario. This is the TDD spec source.
  3. Pin the bug as the TDD feature — set .specify/feature.json so feature_directory points at BUG_DIR (e.g. .specify/bugs/<slug>). Everything downstream (tdd.plan, tdd.run, tdd.verify) then operates on this directory.
  4. Plan the tests — run /speckit.tdd.plan to derive BUG_DIR/tdd/test-list.md. If no plan.md exists in BUG_DIR, use tdd.plan outer-only.
  5. Drive the loop — run /speckit.tdd.run all. It writes the failing reproduction test, proves it RED, applies the minimal fix, proves it GREEN, refactors while green, and logs evidence in BUG_DIR/tdd/cycle-log.md.
  6. Implement non-behavior work — run /speckit.implement for any scaffolding, config, or wiring tasks the loop left for it.
  7. Write the fix report (Execution step 4), referencing the tdd/ artifacts.

In TDD mode the "Apply the remediation" step below is performed by tdd.run, not by hand — do not pre-write the implementation before the test is red.

Execution

  1. Confirm the plan
    • Restate, in 3–6 bullets, what you are about to change and where, based on the assessment.
    • If the assessment's verdict is invalid, stop — there is nothing to fix. Tell the user and exit.
    • If the verdict is likely valid, needs reproduction and there are unresolved [NEEDS CLARIFICATION] items, flag them and ask the user whether to proceed in interactive mode, or stop in automated mode.

Optional — isolate the fix on a branch

By default the fix is applied to the current branch. To match how /speckit-specify isolates feature work, you may ask bug.fix to create a dedicated branch (or git worktree) first:

  • Parse the user input for branch / --branch / worktree / --worktree (or branch=true / worktree=true). These are mutually exclusive; prefer --branch unless the user explicitly asks for a worktree.
  • Determine the branch name <prefix>/<BUG_SLUG>, where <prefix> comes from .specify/extensions/bug/bug-config.yml (branch_prefix, default fix). Example: fix/login-timeout. If a branch with that name already exists, stop and ask the user how to proceed (reuse it, choose another name, or skip isolation).
  • Branch mode (--branch): run git checkout -b <prefix>/<BUG_SLUG> from the current branch (assumed clean or committed).
  • Worktree mode (--worktree): run git worktree add ../<repo>-<BUG_SLUG> -b <prefix>/<BUG_SLUG> so the fix lives in a separate working directory; then continue operations there.
  • If Git is unavailable or the directory is not a Git repository, skip isolation with a warning and apply the fix on the current branch.
  • State which mode was used in your reply; all subsequent edits happen on that branch/worktree.
  1. Apply the remediation

    • TDD mode: if tdd_enabled is true, you already drove the fix via tdd.run in the TDD-mode section above. Skip to step 3 and step 4, and reference the tdd/ artifacts (test-list, cycle-log, verification) in the report.
    • Make the code changes described by the preferred remediation. Stay within the files listed by the assessment unless newly discovered evidence requires expanding scope (in which case, log the expansion explicitly in the report).
    • Add or update the tests called out in the assessment so the bug cannot regress silently.
    • Keep the change minimal — do not refactor unrelated code, do not introduce dependencies that the assessment did not call for.
    • If you discover the assessment was wrong (the proposed fix does not work, the root cause is elsewhere), STOP modifying code, document the new finding in the fix report under Deviations from Assessment, and recommend re-running /speckit-bug-assess.
  2. Run local checks

    • If the project has obvious test commands (e.g., pytest, npm test, cargo test), run the tests that exercise the changed paths. Capture pass/fail and key output.
    • Do not run destructive or network-dependent suites without the user's consent.
  3. Write the fix report

    Write to BUG_DIR/fix.md using this structure:

    # Bug Fix: <short title>
    
    - **Slug**: <BUG_SLUG>
    - **Fixed**: <ISO 8601 date>
    - **Assessment**: ./assessment.md
    - **Status**: applied | partial | not-applied
    - **TDD artifacts**: ./tdd/test-list.md, ./tdd/cycle-log.md, ./tdd/verification.md (present only when the fix ran in TDD mode)
    
    ## Summary
    
    <One or two sentences describing what was changed and why.>
    
    ## Changes
    
    | File | Change | Notes |
    |------|--------|-------|
    | `path/to/file.py` | <added / modified / removed> | <short note> |
    | `path/to/test_file.py` | added test | <short note> |
    
    ## Diff Highlights (optional)
    
    <Short, illustrative snippets of the most important hunks — not a full diff dump.>
    
    ## Tests Added or Updated
    
    - `path/to/test_file.py::test_name` — <what it pins down>
    
    ## Local Verification
    
    - Commands run: `<command>` → <result, brief>
    - Manual checks: <what was verified by hand, if anything>
    
    ## Deviations from Assessment
    
    <Empty if none. Otherwise, list any places where the actual fix departed from the proposed remediation and why.>
    
    ## Follow-ups
    
    - <suggested cleanup, monitoring, doc update, etc.>
    
  4. Report back with:

    • The slug and BUG_DIR/fix.md path.
    • The status (applied, partial, not-applied).
    • Which branch/worktree the fix was applied to (or "current branch" if isolation was not used).
    • The next suggested step(s), in order:
      • /speckit-bug-pr slug=<BUG_SLUG> (open a PR from the fix branch, linking the issue).
      • Then: /speckit-bug-test slug=<BUG_SLUG> (validate the fix). In TDD mode this runs tdd.verify against the bug directory and reports its verdict.

Guardrails

  • Never modify files outside the project workspace.
  • Never edit assessment.md — it is the contract you are working against. Record disagreements in fix.md under Deviations from Assessment.
  • Never delete files unless the assessment explicitly required it.
  • Never overwrite an existing fix.md without confirmation.

04 同仓库更多 Skills