strategic-implementation

Plugin strategic-implementation in wayneliuq/claude-skills. Commands read from the repo's manifest — 18 days ago.

Skills (11)

• clarify

  Entry gate for strategic-implementation v2. Collects all document references upfront (architecture, UX/PMF, security, schema), asks only approach-changing questions, and captures the autonomy level. One pass — no follow-up rounds.

• executing-plans

  Executes an approved v2 execution plan, one deliverable at a time. Deliverable-gated (not session-based). Per-deliverable pre-flight env check, validation per declared method (preview / cli / tdd / post-hoc), atomic commit per deliverable, deviation log. Invokes post-execution for regression-check o

• execution-plan

  Drafts the execution plan for an approved product brief inside Claude Code plan mode, invokes the review skill to critique the draft, patches the plan with review findings, and exits plan mode on PM approval. The execution plan — not the brief — is what the agent panel reviews.

• learnings-benchmark

  Explicit-invocation A/B benchmark of project learnings vs control on a pinned model. Branch-only execution; isolated worktree per trial; blind judge scores outputs; per-learning keep / retire / modify report committed to the benchmarking branch. Run infrequently to decide which learnings still earn

• post-execution

  Post-execution skill with three modes — regression-check (auto after all deliverables complete), triage (PM-reported behavioral issue), learnings-synthesis (validation-log accumulation).

• product-brief-drafter

  Drafts or revises the single PM-facing artifact for a feature — the product brief. Two modes, draft (fresh) and revise (diff-based against PM inline feedback). The brief is the only document the PM approves before execution; it is also the input to execution-plan.

• prune-tests

  Explicit user-invoked skill. Removes line-level unit tests that were intentionally kept sparse pre-GA. Gated by the GA state flag — refuses to run post-GA. Preserves acceptance/goal-level tests. Per-batch PM confirmation before deletion.

• review

  Tiered review orchestrator. Runs pre-filter + generalists (alignment always; plan-simplify unless the plan is trivial; user-validation unless the plan has zero user-observable surface) in parallel on the execution plan, then invokes specialists only on dimensions flagged by a generalist or matched b

• simplify

  Code-level simplicity reviewer. Scans changed files (default `git diff <merge-base>...HEAD`) for reuse misses, dead code candidates, comment hygiene problems, and shape/naming inconsistencies. Reports only — never auto-edits source. Graph-first scan via code-review-graph MCP; file reads only as fall

• strategic-implementation

  Outcome-first orchestrator. Routes PM-described work outside-in — clarify → product brief → optional UI mockup → execution plan (in plan mode) → deliverable-gated execution → post-execution. Single-artifact approval (the brief). Plan mode is both drafting environment and approval gate for the execut

• ui-mockup

  Generates a static HTML mockup as the visual contract for non-trivial UI work. Sits between brief approval and execution-plan drafting. One file, inline CSS, no JS framework, no build. Reuses repo design tokens. PM iterates via conversational feedback in chat. Mockup-vs-brief conflicts route back to

Subagents (10)

• alignment

  Generalist reviewer. Checks the execution plan against the approved product brief, the project's architecture document, and product/UX goals. Owns the first-pass tier — flags which specialist dimensions need deeper review.

• boundaries

  Specialist reviewer for system boundaries — security (attack surface, access control, secrets), data model (schema correctness, migration safety, backwards compatibility), and API contract (interface completeness, versioning, error contract).

• frontend-engineer

  Contextual specialist for UI/UX correctness. Launched only when `alignment` or the pre-filter flags that the plan touches UI components, pages, user-facing flows, or interactive state. Absorbs UX review — interaction patterns, accessibility, empty/error states, and user-journey completeness.

• frontend-quality

  Shared frontend taste reference for strategic-implementation. Two tiers — SLIM (legibility/IA, applied to throwaway mockups) and RICH (pixel quality, applied to shipped UI). Owned in-plugin; not a separate skill. Referenced by ui-mockup (slim) and agents/frontend-engineer (rich).

• plan-gate

  Tightly-scoped gate agent that filters proposed execution-plan patches against the brief's deliverables and success signal. Default use is post-specialist scope discipline; designed to host additional post-plan gate checks via `mode`.

• plan-simplify

  Single-pass simplicity reviewer. Runs in parallel with `alignment` against the execution plan and looks for a shorter path to the brief's success signal and per-deliverable validation. Returns either PASS (with minor flags) or ALTERNATIVE (with a described shorter path).

• runtime-risk

  Specialist reviewer for runtime behavior under load and for supply-chain risk — performance (queries, caching, blocking operations, unbounded resources) and dependencies (necessity, license, maintenance, version pinning).

• technical-expert

  Specialist reviewer for implementation-level correctness. Identifies libraries, frameworks, and runtimes in the execution plan, surfaces known pitfalls, and catches step-ordering errors or missing integration steps. Does not review tests, boundaries, or runtime risk — those are owned by other specia

• tests

  Specialist reviewer for validation adequacy. The brief carries user-facing acceptance steps (outcome-only); the execution plan picks the implementation method (preview / CLI / TDD / integration-test / post-hoc). This agent checks that each deliverable's chosen method honestly demonstrates the brief'

• user-validation

  Generalist reviewer. Stages the brief's named target user and walks every user-acceptance step against the execution plan to find where the user-visible path breaks. Owns PMF, walkthrough, and end-to-end reachability — including the "client-built, pipeline-missing" loophole where UI deliverables loo

Hooks (5)

• SessionStart

  ${CLAUDE_PLUGIN_ROOT}/scripts/store/cache.sh session --background

• PostToolUse

  ${CLAUDE_PLUGIN_ROOT}/scripts/hook-bash-counter.sh

• PostToolUseFailure

  ${CLAUDE_PLUGIN_ROOT}/scripts/hook-error-tracker.sh

• FileChanged

  ${CLAUDE_PLUGIN_ROOT}/scripts/hook-validation-log-watcher.sh

• Stop

  ${CLAUDE_PLUGIN_ROOT}/scripts/hook-stop-orchestrator.sh