Install
$ agentstack add skill-neplich-dev-agent-skills-feature-implementor ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
Feature Implementor
Implements confirmed Engineer TRDs through a durable docs/engineer/{feature_path}/IMPLEMENTATION_PLAN.md, then verifies and closes out the plan. This entry file owns the gates that must run immediately when the skill is triggered; execution details live in _internal/.
Before plan confirmation, load only _internal/planner/INSTRUCTIONS.md plus _internal/_shared/coding-rules.md. After the user confirms the exact plan, load _internal/implementor/INSTRUCTIONS.md. For self-review and closeout, load _internal/reviewer/INSTRUCTIONS.md and _internal/_shared/output-conventions.md.
PM Handoff Entry Gate
Do not execute implementation, write code, or create a plan unless the request has an explicit PM handoff packet for Engineer / delivery-adjacent Engineer work, or an equivalent confirmed document chain: docs/pm/{feature_path}/PRD.md, docs/engineer/{feature_path}/TRD.md, and a current implementation-scope decision.
The PM-side packet field definition lives in agents/product_manager/skills/idea-to-spec/_internal/_shared/skill-map.md. If the user directly asks this skill to "implement", "build", "change a feature", or "do the code" without PM handoff or equivalent confirmed docs, stop and return the request to pm-agent for classification. Direct invocation does not bypass this gate.
When To Use
- Implement a feature or behavior change already covered by confirmed PM scope
and Engineer TRD.
- Turn a confirmed TRD into an implementation plan, code changes, tests, and
verification evidence.
- Implement spec-backed bug fixes only after Engineer/debugger routing confirms
the fix is implementation work against approved PRD/TRD behavior.
Do not use this skill for:
- greenfield product ideas or raw feature requests without PM scope
- behavior changes that need PRD or product decision updates first
- missing, stale, or mismatched TRDs
- bug investigation before expected behavior is aligned
- tests-only or delivery-only work
PRD And TRD Alignment Gate
Before writing or updating IMPLEMENTATION_PLAN.md, resolve the canonical feature_path and read docs/pm/{feature_path}/PRD.md, docs/engineer/{feature_path}/TRD.md, and docs/pm/{feature_path}/DECISIONS.md or equivalent product decisions when present.
Proceed only when the PRD exists, the TRD exists, their feature_path, parent_feature, and feature_level metadata match, and the TRD related_prd points to docs/pm/{feature_path}/PRD.md. Legacy single-level PM docs without feature-path metadata may be read as level-1 features, but new or changed plans must write explicit feature-path metadata.
Classify before planning: approved scope proceeds; expectation changes return to pm-agent:idea-to-spec using existing-project-update; missing PRD or decisions return to PM; stale, incomplete, path-mismatched, or conflicting TRDs return to engineer-agent:trd-gen; requests to skip alignment are blockers or risks, not implementation permission.
The TRD gap packet must name the missing technical decisions, affected components, data/API/integration impacts, validation commands, rollout risks, and the boundary: the finder only clarifies gaps; trd-gen completes the TRD.
UI Design Handoff Gate
For frontend UI, interaction, visual, component, usability, or information hierarchy changes, check docs/design/{feature_path}/ui-ux-spec.md and docs/design/{feature_path}/visual-system.md.
Cite existing design docs when they cover the change. If the change is narrow enough not to require Designer updates, state why in the implementation plan. If design inputs are missing, stale, or conflicting, stop before planning and hand the gap through engineer-agent to designer-agent.
Plan And Archive Gate
docs/engineer/{feature_path}/TRD.md is the technical input contract. IMPLEMENTATION_PLAN.md maps that TRD to concrete files, sequence, delegation, verification, and closeout. It must not rewrite PM scope or TRD decisions.
Before creating or replacing an active plan, scan docs/engineer/{feature_path}/IMPLEMENTATION_PLAN.md and its implementation-plans/archive/ directory.
If an active plan exists and no handling decision is recorded, ask the user to choose exactly one: archive completed plan then create new active plan, continue updating the current plan with a version bump, or archive the old plan as Superseded with a reason then create a new active plan.
Plan form strength follows change_tier from the PM handoff or AGENTS.md. hotfix may use the lightweight plan form allowed by the repository contract; standard and major keep the full plan confirmation flow. Tiering never waives PRD/TRD alignment or evidence.
Archival happens only after closeout and user/maintainer approval. Archive paths must use docs/engineer/{feature_path}/implementation-plans/archive/IMPLEMENTATION_PLAN-.md. The active plan path remains fixed at docs/engineer/{feature_path}/IMPLEMENTATION_PLAN.md.
Implementation Flow
- Gather PRD, DECISIONS, TRD, relevant design docs, repo structure, and active
plan/archive state.
- Write or update the plan with file list, order, verification, alignment
result, feature metadata, and sub-agent split decision.
- Present the exact plan and wait for user confirmation.
- Implement only the confirmed scope, reading each file before editing.
- Verify with deterministic commands.
- Self-review against PRD/TRD/design docs, repo rules, security basics, and
unrelated-change safety.
- Update closeout before delivery.
For complex multi-file or spec-heavy work, use separate implementation and validation sub-agents when available. Do not use that split for small edits or when the user opts out; the plan gate still applies.
Closeout And QA Handoff
After implementation and deterministic checks, update or confirm closeout. Closeout records final status, changed files, commands run or skipped, skill / fresh subagent eval status when applicable, residual risks, and next owner.
If frontmatter says status: "Implemented", the body must not keep unresolved planning wording such as "waiting for confirmation", "not started", or "pending execution" except as clearly historical context with a resolved result. Runtime eval artifacts must not be committed.
When user-facing flows, acceptance paths, permissions, login, data setup, or regression coverage may be affected, produce a QA E2E handoff package. Do not create QA E2E cases unless explicitly routed to QA work.
Key Principles
- Read before write.
- Keep changes scoped to the confirmed plan.
- Prefer existing project conventions over new abstractions.
- Every implementation decision traces to PM/TRD/design docs.
- Do not perform missing PM, Designer, QA, DevOps, or Security responsibilities.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: Neplich
- Source: Neplich/dev-agent-skills
- License: Apache-2.0
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.