implement-plan
/implement-plan
Execute an approved implementation plan phase by phase with continuous progress until stop conditions. Use only when the user asks to implement an approved plan, build a named plan, or execute a plan they identify. Not for small unplanned direct work, exploratory fixes without a plan, or creating a new plan.
- Author
- joshgons ↗
- Repository
- agent-skills ↗
- Licence
- MIT
- Last repository change
- 09.11.2026
Assessment and security
- Files in this package
- 4
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 09.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Execute an approved implementation plan phase by phase with continuous progress until stop conditions. Use only when the user asks to implement an approved plan, build a named plan, or execute a plan they identify. Not for small unplanned direct work, exploratory fixes without a plan, or creating a new plan.
Original author description
Execute an approved implementation plan phase by phase with continuous progress until stop conditions. Use only when the user asks to implement an approved plan, build a named plan, or execute a plan they identify. Not for small unplanned direct work, exploratory fixes without a plan, or creating a new plan.
How it works
- Select the plan
- Read first
- Preflight
- Escalation — STOP vs decide
- Execution loop
- Finish
SKILL.md preview
--- name: implement-plan description: "Execute an approved implementation plan phase by phase with continuous progress until stop conditions. Use only when the user asks to implement an approved plan, build a named plan, or execute a plan they identify. Not for small unplanned direct work, exploratory fixes without a plan, or creating a new plan." --- # Implement Plan Execute a selected approved plan as a continuous loop. Do not turn small unplanned direct work into a planning blocker—decline this skill when the user wants a quick fix without a plan. ## Select the plan 1. If the user names or tags a plan, use that plan. 2. If the repository uses a plan lifecycle (folders or status fields), verify the selected plan is **approved** / ready before starting. 3. If multiple approved candidates exist and the user did not choose, **ask which plan** — never silently pick the oldest. 4. If no plan exists and the user asked to implement a plan, stop and say planning is required first. If they asked for unplanned direct work, do not invoke planning as a gate. ## Read first - The selected plan in full (non-goals, risks, blind spots, verification — not only the implementation section) - [references/execution-lifecycle.md](references/execution-lifecycle.md) - [references/verification-strategy.md](references/verification-strategy.md) - Relevant current source for every file the plan names ## Preflight 1. Inspect workspace hygiene with path-aware status; preserve unrelated workstreams. 2. Spot-check plan assumptions (paths, exports, types, behaviors) against current source. 3. Check recent history on targeted files for renames/refactors that invalidate the plan. 4. Reconcile stale plan steps and material architecture drift **before** coding; update plan docs when local lifecycle conventions exist. 5. Discover validation commands from the repository (`package.json`, `Makefile`, `AGENTS.md`, CI config) — do not assume a package manager or script names. ## Escalation — STOP vs decide **STOP and surface:** - Baseline checks already broken before your edits (when a baseline check is defined and runnable) - Named path missing or identifier/signature mismatch vs the plan - Load-bearing assumption fails verification - Literal plan execution would break working behavior - Unrelated substantial bug found mid-flight (note it; do not expand scope) - A failing behavior cannot be explained with a focused, red-capable signal — hand off to a focused debugging workflow (`$diagnose-bugs`, if installed) before speculative fixes - User contradicted the plan after it was written - Destructive operations (delete data, reset, drop, force-push) appear necessary **Decide and proceed:** naming, formatting, import order, exact placement within a stated region, obvious refactors of lines already being touched. ## Execution loop While the next phase or titled sub-step is actionable: 1. Reconcile the step against current code (and architecture anchors if the repo has them). 2. Read each file fully before editing; prefer small scoped edits; do not invent scope. 3. Implement the durable solution for the current phase — continue without waiting for a nudge when the next step is defined and actionable. 4. Run the narrowest repository-discovered validation for the change; scale up for broad/risky phases per [references/verification-strategy.md](references/verification-strategy.md). 5. Update plan status / lifecycle / next-action **only when those conventions exist**. 6. For long or multi-phase work, take periodic maintenance checkpoints: reconcile the plan with current code, remove temporary layers, re-check priorities, and run a focused self-review before drift accumulates. 7. Proceed to the next phase unless a stop condition applies. ## Finish 1. Verify each acceptance criterion individually; classify unperformed manual checks explicitly. 2. Self-review the **path-scoped** diff; fix only issues introduced by this workstream. 3. Apply documentation follow-ups the plan requires when still warranted. 4. **Learning assessment** — when compound-engineering is in play or material reusable knowledge emerged, assess candidates for `$capture-learning` (create/update/supersede/skip) and note any stale guidance that warrants `$refresh-learnings` (both if installed). Do not write learnings routinely from this skill; do not duplicate capture/refresh workflows here. 5. Never commit, push, deploy, migrate, or destructively mutate without explicit authorization. 6. Report: files changed; checks run and results; acceptance criteria status; manual checks remaining; risks; learning assessment decision; exact plan path. Delegated adversarial review is optional and only when the user asks — recommend `$use-codex` or `$use-cursor` advisory workflows if those skills are available; do not require them. Do not automatically fan out external models.
Files in this package (4)
- SKILL.md 5 KB
- agents/openai.yaml 1 KB
- references/execution-lifecycle.md 2 KB
- references/verification-strategy.md 2 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · joshgons/agent-skills | 90 | MIT |
| tzachbon/srulik-toolkit | 100 | MIT |
| atomicinnovation/accelerator | 100 | MIT |
| patrickdappollonio/claude-plugins | 100 | MIT |
| paultyng/skill-issue | 100 | MIT |
| mhylle/claude-skills-collection | 95 | MIT |
Database snapshot: 13.11.2026 · package version f7217e75307e (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90implement
Execute an implementation plan step by step with verification checkpoints and progress tracking. Use when the user asks to "implement the plan", "execute the steps", "start implementing", or has an approved plan ready to implement. - 95sprint-executor
Execute approved sprint plans with test-driven development, continuous linting, progress tracking, and pause points. Supports parallel milestone execution via Task sub-agents. Use when user says "execute sprint", "start sprint", or wants to implement an approved sprint plan. - 90rpi-implement
Execute an approved plan from .rpi/plans phase by phase with TDD and per-phase verification. Use when user says 'implement the plan', 'start implementing', 'run the next phase', or just approved a plan. - 90rpi-implement
Execute an approved plan from .rpi/plans phase by phase with TDD and per-phase verification. Use when user says 'implement the plan', 'start implementing', 'run the next phase', or just approved a plan. - 95sprint-executor
Execute approved sprint plans with test-driven development, continuous linting, progress tracking, and pause points. Supports parallel milestone execution via Task sub-agents. Use when user says "execute sprint", "start sprint", or wants to implement an approved sprint plan. - 95Implementation
Execute implementation tasks from task documents following TDD workflow, phase-by-phase execution, and real-time progress tracking. USE WHEN implement, execute tasks, run tasks, start implementation, build feature, execute implementation, code the feature.