| name | cursorqb-implementer |
| description | Gated, goal-backed Step 4 of the CursorQB workflow - implement one bounded, reversible, testable slice from an audited plan. Use after the Step 3 audit when the user approves implementation, or via /cursorqb-implement. Runs only when the audit is PASS or PASS_WITH_WARNINGS with no P0/P1 findings; otherwise it stops and recommends repair. Selects a single READY sub-plan, defines the validation command first, makes a minimal change, and verifies before claiming done. |
CursorQB Implementer (Step 4, gated, goal-backed)
Implement one bounded, reversible slice from the audited CursorQB plan. This is the only
CursorQB step that changes source code, and it runs strictly behind the audit gate. Read
the full Step-4 spec in fourth-planner.md (next to this skill) and follow it end to end.
Execution model
- In-session, goal-backed. Read the bundled
fourth-planner.md spec and follow it. Register the slice goal via the Cursor define-goal skill (get_goal then create_goal); if the goal tool is unavailable, proceed under the in-context goal contract - behavior is identical.
- Read the reliability notes. Resolve the plugin root by walking up to the folder containing
.cursor-plugin/plugin.json, then read references/workflow-quality.md (especially Step 4 Token Discipline) before starting.
- Scope is the user's workspace (cwd). All reads/edits target the user's active workspace, never this plugin directory.
Gate (mandatory, before any change)
Run the bundled validator in Step 4 mode (resolve <plugin-root> as above):
python3 <plugin-root>/scripts/validate_planner_docs.py --root . --mode step4
BLOCKED audit status -> do not implement; surface the blocker and minimal unblock action; stop.
- Any P0 or P1 finding -> do not implement; recommend the targeted repairs first (
cursorqb-auditor fix list + cursorqb-planner repair loop); stop.
- PASS, or PASS_WITH_WARNINGS with only P2/P3 -> proceed and keep the warnings visible.
python3 unavailable -> read the audit's ## 1, ## 12, ## 13, ## 15 sections and apply the same gate manually.
Token discipline
- Read
Planner-docs/Sub-Planing-Audit.md and Planner-docs/Sub-Planing-Index.md first.
- Select ONE
READY or READY_WITH_WARNINGS sub-plan from the audit's ## 12. Step 4 Readiness Assessment table (highest-priority first, unless the user names another).
- Load only that sub-plan and the repo files needed for the slice. Do not load all sub-plans.
Slice procedure
- Read
git status, README.md, AGENTS.md, Makefile, the audit, and the selected sub-plan.
- Determine the validation/test command FIRST; prefer existing repo commands over invented ones; write a failing test first when feasible.
- Make the minimal, reversible change for the slice.
- Run the focused tests plus the relevant
make smoke/check target.
- Verify with fresh evidence before claiming done.
- Report an exact blocker or success; separate code-delivery status from external config/credential blockers.
Leverage installed skills (optional, with fallback)
If installed/available, use these by scope; if not installed, do not stop - continue with the audit, the
selected sub-plan, repo instructions, and existing validation commands using the same principles:
- Superpowers
executing-plans or subagent-driven-development for execution structure.
- Superpowers
test-driven-development for code changes.
- Superpowers
verification-before-completion before asserting the slice is done.
- The security review (
review-security / security-review) for security-, policy-, secret-, or command-execution-sensitive changes.
Stop rules
- One sub-plan and one reversible slice per run.
- Never write secrets, tokens, private keys, or local credentials.
- Do not commit, push, open a PR, deploy, or mutate external systems unless the user explicitly asks in this run.
- Do not claim success without running the validation command and confirming its output.
Output
Report: which sub-plan slice was implemented, the validation command and its result,
files changed, remaining P2/P3 warnings to keep visible, and the recommended next slice
or the exact blocker. After a successful slice, the user can run /cursorqb-implement
again for the next READY sub-plan.