marketplace

Document-assembly, pre-write invariant-guard, and atomic-write algorithm for forgeplan-map-pack's EMIT stage (RFC-023 Proposed Direction SS1 roster row 7 + Invariant #1; PRD-075 FR-5; SPEC-003 SS C0 the one-sentence contract, SS C3 gate G4). Covers: assembling meta/canvas/composition/zones/nodes/edges from the extraction+edges+composition inputs, the pre-write self-check trio RFC-023's own function signature assigns to this stage (no zone-cell overlap, every edge endpoint in nodes, every node.zone in zones -- independently re-derived afterward by the guardian's GC-2, NOT to be confused with SPEC-003's differently-numbered C1 INV-1/2/3 trio which this stage satisfies by disciplined passthrough construction instead), the reject-own-output-before-write rule, the exact tmp-rename atomic-write recipe mirroring map-guardian.mjs's own write, and the status:proposed + NEEDS_CONFIRM sentinel contract. Invoked by the map-emitter agent only. Triggers: "assemble map.json", "invariant guard before write", "atomic tmp-rena

Zone/node/mega-node extraction algorithm for forgeplan-map-pack's EXTRACT stage (RFC-023 Proposed Direction SS1 roster row 5, "THE HEART"; PRD-075 FR-3; SPEC-003 SS C1 INV-2, SS D4). Covers: the merge-then-dedup rule across the 3 scan scratch files, the content-hash node-id formula (sha1(kind+":"+path_or_slug)[:12], INV-2 -- never a label or counter), zone_hints binning with the z.core default-home fallback, why zone.cols is pinned from the composition and never derived from node count (append-stability), the GROUPED (per-kind) mega-node collapse rule for over-capacity zones (never one opaque zone-wide dump), and RU narration attachment into description_ru from real docs only. Invoked by the zone-extractor agent only -- not a general-purpose codebase-summarization skill. Triggers: "mint node id", "content-hash id formula", "zone binning", "mega-node collapse", "pin zone cols", "merge scan scratch files", "description_ru narration".

Edge trust-classification and grep-verification algorithm for forgeplan-map-pack's VERIFY stage (RFC-023 Proposed Direction SS1 roster row 6; PRD-075 FR-4; SPEC-003 SS D2 the 11 VALID_RELATIONS, SS D3 the typed-link/code-dep namespace default rule). Covers: the namespace classification rule, the exact 11-relation allowlist for typed-link edges, the mandatory Bash grep-verification pass for code-dep edges with the verified_by="grep:<pattern>" format map-guardian.mjs's XC-2 re-runs verbatim, the unconditional drop-if-unverified rule (never emit a code-dep edge as noise), and argv-safe grep discipline (no shell interpolation of untrusted pattern content). Invoked by the edge-verifier agent only. Triggers: "classify edge namespace", "verify code-dep edge", "grep verification pass", "11 valid relations", "drop unverified edge", "verified_by format".

Forgeplan structured engineering workflow — plan features, create PRD/RFC/ADR, check project health, route tasks to correct depth, review artifacts, think before coding. Triggers: forge, forgeplan, route task, create prd, check health, review artifact, activate, lifecycle, R_eff, evidence. Not for: simple bug fixes, formatting, trivial changes.

Live reference for the 66 mcp__forgeplan__* MCP tools and the 82 forgeplan CLI subcommands shipped in forgeplan 0.32.1. Agents and humans consult this skill when they know what they want to do (create an artifact, score a graph, dispatch sub-agents, detect anomalies) but don't yet know which tool name to call. Organised by use-case (lifecycle / discovery / quality / health / AI-driven / multi-agent / brownfield / activity / FPF / playbooks / plugins / release / safety) rather than alphabetically. Each section maps tools to canonical use-cases + which Profile A / B / C / D agent or which skill should call them. Includes known safety warnings (forgeplan#350 MCP body literal-string-only, forgeplan#348 score-all hint typo, forgeplan#351 plugins doctor install syntax). Triggers: "forgeplan tool reference", "which forgeplan command", "forgeplan-cookbook", "/forgeplan-cookbook", "what mcp tool", "how do I create artifact via mcp", "forgeplan cli vs mcp", "forgeplan capabilities"

Task-planning skill. Takes a SPECIFIC TASK as input (e.g., "refactor auth module", "ship payment service", "audit security of API gateway"), classifies it into one of 14 routing-map contexts, applies the matching methodology, and renders a structured Plan using templates/smith-plan.md. Dispatches the `smith` agent for routing decisions, returns a Plan markdown ready for orchestrator hand-off. Optionally creates a NOTE artifact to persist the Plan via forgeplan. Triggers: "smith plan", "/smith-plan", "спланируй задачу", "plan this task", "how should we approach", "как подойти к задаче"

Educational routing skill. Walks the user through smith's methodology routing matrix without committing to a specific task. Use to learn "when do I apply BMAD vs SPARC?", "what methodology for brownfield?", "explain the bug fix workflow". Pulls from routing-map.md (14 contexts + 29 methodology cards) and sections/NN-*.md (detailed playbooks). Output: educational comparison + recommendation when user describes their situation. Does NOT produce a Plan artifact — for that, use /smith-plan. Triggers: "smith routing", "/smith-routing", "explain methodology", "какая методология", "when do I use", "compare methodologies", "routing walkthrough", "BMAD vs SPARC", "Strangler Fig or rewrite", "show me the routing matrix"

Master-orchestrator entry-point. Reads project state (forgeplan_health, memory_recall, git), classifies context (1 of 14: greenfield / brownfield / feature / bug-fix / refactor / decision / audit / discovery / tech-debt / incident), applies methodology routing matrix, returns a structured Plan and recommends specialist-agent dispatches. Calls the `smith` agent for the planning + the routing-map.md for the brain. Default behaviour: status + recommend next step. Sub-modes via args: `bootstrap` (greenfield onboarding), `plan <task>` (per-task plan), `routing` (educational walkthrough), `status` (current state only). Triggers: "smith", "/smith", "кузнец", "что дальше", "куда идём", "возьми управление", "scrum master", "master orchestrator", "captain mode", "оркеструй", "take charge"

Runs the implementation team from a finished research report — reads IMPLEMENTATION-PLAN.md (phases, file ownership, acceptance criteria) and the SUMMARY with ADRs, spawns teammates by phase (sequential or parallel), and verifies the result via build/typecheck/tests. On completion, **automatically writes a single EvidencePack via forgeplan MCP** (`mcp__forgeplan__forgeplan_new` + `forgeplan_update` + `forgeplan_link`) so the build lands in the artifact graph — no manual step required (CLI fallback if MCP not connected; explicit warning if neither available, never a silent skip). Use when research is already done (via research or manually) and only "execute the plan" is left. Triggers (EN/RU) — "build from research", "execute IMPLEMENTATION-PLAN", "implement based on research report", "имплементируй по research", "выполни план", "/build".

Meta-orchestrator — takes a natural-language task, classifies it (research / docs / feature / review / bug / refactor / status), builds a pipeline from other skills (research → write-doc → wave-sprint → audit), shows the plan, and executes with approval checkpoints. Use when the user wants "do X" without naming a specific skill. **MCP-first by delegation** (per PRD-022 Tier B): does not call forgeplan directly — delegates to sub-skills (research, refine, sprint, audit, briefing) that are themselves MCP-first with CLI fallback; the orchestration layer is forgeplan-agnostic. Triggers (EN/RU) — "do X", "implement and document", "research and write RFC", "сделай", "разберись и реализуй", "проведи ревью ветки", "/do".

Wave-based execution of a feature/task by a multi-agent team — research → wave plan (5-8 agents in 2-5 waves) → approval → wave-by-wave teammate spawn via TeamCreate. Two modes — full sprint (with research phase) and lightweight wave (uses current chat context). Each wave is independent parallel work by agents under strict file ownership; dependent tasks go in the next wave. At every wave-close, **automatically emits one EvidencePack per touched artifact via forgeplan MCP** (`mcp__forgeplan__forgeplan_new` + `forgeplan_update` + `forgeplan_link`) — no opt-in. At sprint-close, prompts the user to bulk-activate the collected EVIDs. Soft CLI fallback if MCP unavailable; warns instead of silent-skip if a sprint closes with zero EVIDs. Use for large feature implementation, refactor sprints, milestones. Triggers (EN/RU) — "sprint", "wave plan", "implement feature in waves", "запусти спринт", "распланируй волны", "реализуй фичу", "implement RFC-XXX", "/sprint", "/wave".

CANVAS DS-build conventions knowledge base — the rule set the canvas-guardian audits a Pencil Design System against (HOW it was built, not whether it matches requirements). Single-source refs/slots/no-detach, $--token usage + Category/Variant naming, atomic layering + screens-not-reusable, and snapshot-layout health (clipping/spacing). Use when auditing a DS snapshot, running /canvas-audit, or learning a single DS convention. Triggers on: ds audit, design system conventions, canvas guardian, ref vs detach, reusable component, token usage, atomic layering, screens not reusable, clipping, height-aware spacing, snapshot_layout, Category/Variant naming, проверь дизайн-систему, конвенции DS, аудит дизайн-системы.

Pencil designer knowledge base for the CANVAS methodology — how to design and extend a .pen Design System with refs, slots, descendants, Layout B, atomic organization, themed tokens, style presets, and UX-law-driven node constraints. Use when designing or extending a Pencil DS, building screens/components, fixing layout clipping, organizing ATOMS/MOLECULES/ORGANISMS/TEMPLATES, wiring theme variables, or translating UX laws into Pencil nodes. Triggers on: pencil design, design in pencil, .pen, design system, batch_design, refs and descendants, Layout B, subsidebar, atomic design, design tokens, theme variables, screen template, table actions, clipping, snapshot_layout, get_screenshot, choosing and recording a brand/style, warm-paper + Expo monochrome shown as worked examples, style guide, ds organization, ux constraints in pencil.

CANVAS porting playbook — Pencil design system -> Storybook -> native framework code. Shared contract read by canvas-porter-storybook (Vectorize) and canvas-coder (Assemble); the optional canvas-porter-framework (multi-target port) is dispatched only on an explicit multi-framework request, out of the default pipeline. Step 0 resolves the consuming project's framework as an INPUT — detected from AGENTS.md / CLAUDE.md / package.json, else force-asked — and CANVAS generates natively in that one framework. Lit / Web Components / Shadow DOM are ONE selectable target, used only when the declared stack IS Web Components, not the canon. Covers the token contract (single tokens.json -> CSS custom properties, compiled by the project's token tool — Style-Dictionary is one option), the per-component story spec (variant matrix + slot map + descendant-override points), and the visual oracle (reference screenshots -> visual-regression via Playwright / Storybook test-runner). Use when porting a Pencil DS to tokens/Storybook/

Entry point for the CANVAS design-system -> code pipeline — the master-coordinated walk from a Pencil design to **native code in the consuming project's own framework** (framework resolved via Step 0 — Capture -> Audit -> Norm-check -> Vectorize -> Assemble), an instance of the AD/AID-PDLC sub-cycle contract (ADR-010 / RFC-021), hook-gate=YES. Dispatches the canvas-coordinator master, which walks EVERY phase + EVERY verifier as separate isolated-context Task sub-agents (BMAD-shaped) — including the Pencil-touching Designer and Storybook-Porter (Pencil MCP works in dispatched sub-agents) — with a blocking generator-not-verifier gate at each handoff and a fail-closed PreToolUse tokens-gate that blocks design-system/framework source writes until the tokens RFC is active. EN: Run CANVAS on a design-system slice — Pencil DS -> Audit -> Norm-check vs ForgePlan -> Storybook port manifest (single-source tokens via the project's token tool) -> native component code in the project's own framework + visual tests. Use fo

CANVAS Storybook-gate validation toolkit (RFC-021 FR-4) — the knowledge base read by canvas-storybook-validator (the C4 Storybook-gate, a SUB Task agent) when it certifies the BUILT Storybook against the Pencil source ONLY. Covers the six certifications: (1) story coverage vs the port-manifest variant matrix, (2) visual parity vs the Pencil reference screenshots (visual-regression — Playwright / Chromatic), (3) interaction/play tests (play functions + storybook/test userEvent + expect), (4) structural accessibility via the a11y/axe addon (WCAG), (5) token fidelity (computed styles resolve to the token tool's CSS custom properties, no hardcoded values), (6) coverage thresholds. Documents the Vitest addon (primary, Vite) vs the legacy test-runner (Jest+Playwright fallback), composeStories().run() portable stories, and where this gate sits (generator!=verifier vs canvas-coder — a fresh isolated context). Use when running or reviewing Storybook tests, wiring the validator gate, or writing *.stories.ts test code.

Map a Pencil design system to the ForgePlan source of truth (PRD/ADR/EVID) — the CANVAS Norm-check (N) knowledge base. Use when validating a DS snapshot for coverage (does every required capability have a component, and does every component have a backing requirement) and provenance/traceability (does each component and token trace to an active decision, and do the snapshot tokens match the ADR palette). Produces a requirement->component->artifact traceability matrix + a C4 EVID. Triggers on: norm-check, design system coverage, DS to PRD traceability, requirement to component matrix, scope creep detection, token provenance, gap analysis, DS source of truth, canvas tester.

Scan all active ADR artifacts for fired Revisit Triggers (Evidence Decay). Reports which triggers fired, which need human verification, which are stale because the ADR uses pre-Sprint-Z2 prose format. Use periodically (weekly + at session start) — keeps decisions from quietly going stale. Triggers: "decay watch", "/decay-watch", "проверь триггеры пересмотра", "stale ADR", "revisit trigger check", "evidence decay scan"

Methodology: Layer 4 heal-or-task loop (PRD-074 brick 2 / RFC-020) — propose-not-silent 3-tier dispatch (AUTO / ADI / USER per PRD-032) of the Finding[] that /forge-insight + the insight-watchdog already surfaced. Consumes the frozen RFC-019 Finding contract; never re-detects or re-classifies. EN: Acts on the machine findings recorded by /forge-insight (the NOTE-013 `## Machine findings (auto-tracked)` ledger). Routes each finding onto PRD-032's 3 tiers: AUTO (a hard 2-kind reversible allowlist — phase_mismatch→phase_advance, complete stuck_draft→activate — applied only after a batch confirm + logged + rolled-backable), ADI (ambiguous: an FPF reasoning pass that auto-applies ONLY if high-confidence AND reversible, else escalates), USER (content/irreversible/cross-team: writes a tracked task row + suggests the right marketplace agent, never auto-applies). Default is PROPOSE-and-wait — it NEVER silently mutates the tree (the "AI auto-fixed and broke it" failure PRD-074 exists to prevent); silent-AUTO exists onl

Methodology: insight layer over forgeplan anomaly detection (PRD-074 Layer 2 + don't-forget ledger; RFC-019 brick 1). Deterministic known-noise filter + on-demand semantic digest + idempotent NOTE-013 findings-section writer. Complements the silent SessionStart insight-watchdog hook (same filter, automatic) with a richer manual pass. EN: On-demand deep audit of the forgeplan artifact tree's health. Reads forgeplan_anomalies (+ health + blindspots), filters the two known-noise classes (weakest_link_unresolvable per forgeplan#325 + benign phase_mismatch), ranks survivors by severity, caps at 5, and explains "what each means" — collapsing a 133KB / 189-line raw dump into a few honest observations. For every medium+ finding it upserts a `Kind: finding` row into NOTE-013's dedicated `## Machine findings (auto-tracked)` section (idempotent by finding_id, auto-closes vanished findings, hard cap 20) so real issues are not forgotten. Read-only on the audited tree; the ONLY write is the NOTE-013 findings section (never

Recommends the right canonical agent (or set of agents) for a described engineering task. Consults the `mm-agent-selection` mental model in Hindsight when available, falls back to embedded knowledge of the 17 forgeplan-aware canonical agents + their CRUD-R-A profile matrix (PRD-026). Returns a structured recommendation — primary agent, optional secondaries, profile letter, short rationale, and a ready-to-paste `Task({ subagent_type: ... })` invocation snippet. Use when the user describes a task in natural language and asks "which agent should I use?", or when you (Claude) want to dispatch the right specialist for a phase but are not 100% sure of the canonical mapping. Triggers (EN/RU) — "which agent for", "recommend agent", "what agent should I use", "agent for this task", "какого агента", "посоветуй агента", "/agent-advisor".

Methodology: Cross-marketplace agent/skill suggestion (suggest-only, NO auto-install). EN: When canonical 17 forgeplan-aware agents don't fit the task, searches installed marketplaces (~/.claude/plugins/marketplaces/) for matching agents/skills. Ranks by keyword overlap + trust tier + specificity. Surfaces top 3 with description + install command + security caveats. NEVER auto-installs — user runs the command manually. Falls back to external catalogs (VoltAgent awesome-*, DenisSergeevitch) when no local match. RU: Когда канонических 17 forgeplan-aware агентов не хватает, ищет среди установленных маркетплейсов подходящих агентов/скиллов. Ранжирует по ключевым словам + tier + специфичности. Показывает топ-3 с описанием + командой установки + предупреждениями. НЕ устанавливает автоматически. Fallback на external catalogs. Triggers: "find agent for", "fetch agent", "external agent", "cross-marketplace agent", "search marketplaces", "найди агента", "поищи в маркетплейсах", "/agent-fetcher"

Multi-expert audit of code, architecture, or a finished feature — at least 4 parallel reviewer agents, each with a specialization (logic, architecture/SOLID, typing, security; optionally tests, backend patterns, frontend, task-completion). Each runs its own checklist. Output is cross-validation, score, verdict, and action plan. On completion, **automatically writes an EvidencePack via forgeplan MCP** (`mcp__forgeplan__forgeplan_new` + `forgeplan_update` + `forgeplan_link`) so the audit lands in the artifact graph and R_eff reflects it — no manual step required (CLI fallback if MCP not connected). Use when the user wants a sharp second pair of eyes on code, a PR, a branch diff, or a sprint. Triggers (EN/RU) — "audit", "review", "проверь код", "ревью", "оцени качество", "всё ли правильно", "code audit", "expert review", "/audit".

Autonomous orchestrator for green-light / overnight execution. Takes one task and runs the full engineering cycle (research → plan → sprint → audit → report) end-to-end without approval checkpoints. Uses TeamCreate with explicit file-ownership and blockedBy edges between waves, resolves blockers via ADI (Abduct → Induct → Deduce) instead of asking the user, and only stops on red-line actions (push to main, secret writes, destructive ops, deploys). Use when you want to give one prompt and let it run unattended — overnight runs, bypassPermissions sessions, or when you can't watch checkpoints. Triggers (EN/RU) — "autorun X", "run unattended", "do this overnight", "запусти автопилот", "сделай всё автономно", "ночной прогон", "/autorun".

Deploys the author's Claude Code starter scaffold into a new or existing project — creates a CLAUDE.md from the template, a guides/ folder with author guides (Git Flow, CLAUDE.md best practices), and guides/INDEX.md. Use when the user starts a new project and wants the standard Claude Code baseline, or when adding these guides to an existing project. **No forgeplan dependency** (per PRD-022 Tier B): pure file-scaffolding; if the user wants forgeplan integration after bootstrap, suggest `/fpl-init` (which IS MCP-first). Triggers (EN/RU) — "bootstrap project", "init claude baseline", "install my guides", "scaffold claude code", "поставь мои гайды", "разверни базу", "установи базовый CLAUDE.md", "подготовь проект под claude", "засетапь проект".

Morning briefing on tasks and messages — collects overdue, due today, @mentions, unread chats, and project stats from any available task tracker (Orchestra MCP, Linear MCP, Jira MCP, GitHub Issues) or from local TODO files. When forgeplan is reachable, augments with **MCP-first** artifact signals (`mcp__forgeplan__forgeplan_list` + `forgeplan_blocked` + `forgeplan_stale` + `forgeplan_blindspots`) with CLI fallback when MCP is not connected. Use at the start of the workday or when the user asks "what's on my plate", "morning standup", or similar. Triggers (EN/RU) — "daily briefing", "what's on my plate", "morning standup", "что у меня сегодня", "что висит", "брифинг", "standup", "/briefing".

Interactive C4 architecture diagram generator. Walks through the four C4 levels (Context, Container, Component, Code) via guided questions and produces Mermaid diagrams plus a context document. Pairs with /ddd-decompose (which produces a domain context map) — /c4-diagram adds the deployment / runtime structure layer. Maps cleanly to forgeplan via c4-to-forge.yaml in forgeplan-brownfield-pack. **MCP-first per PRD-022 Tier A** — Epic/Note/PRD/RFC/ADR creation uses `mcp__forgeplan__forgeplan_new` + `_link` + `_reason` when MCP available; CLI fallback when MCP server not connected. Triggers (EN/RU) — "c4 diagram", "architecture diagram", "context container component", "архитектурная диаграмма", "c4 нарисуй", "/c4-diagram".

The OPTIONAL `## Conformance Vectors` enrichment for spec-driven development (ADR-008): a frozen, hash-pinned, language-neutral conformance corpus that is the SOLE behavioral oracle for an implementation, plus a per-language harness and a cross-language equivalence gate. Use when a SPEC targets MORE THAN ONE implementation language (REQUIRED — prose scenarios alone diverge across languages) or for a pure algorithmic core where prose under-specifies. Do NOT use for ordinary single-language features — the SPEC's `#### Scenario` blocks are the oracle there. Proven across 4 languages (TS/Py/Go/Rust) in `reference/semver/` (EVID-119). Triggers: "conformance vectors", "conformance corpus", "cross-language equivalence", "multi-language spec", "frozen oracle", "language-neutral corpus", "spec conformance harness", "конформанс-вектора", "кросс-языковая эквивалентность"

Interview-driven Domain-Driven Design decomposition skill. Walks through identifying bounded contexts, aggregates, ubiquitous language, and domain events for a system or new feature. Outputs a DDD context map (Markdown + Mermaid) plus optional forgeplan artifacts (Epic + per-context PRDs + Spec). Pairs with /fpf-decompose (general decomposition) — use /ddd-decompose when DDD framing matters (multi-team boundaries, complex business domains). **MCP-first per PRD-022 Tier A** — Epic/PRD/Spec/ADR creation uses `mcp__forgeplan__forgeplan_new` + `_link` + `_validate` + `_reason` when MCP available; CLI fallback when MCP server not connected. Triggers (EN/RU) — "ddd decompose", "bounded contexts", "ubiquitous language", "domain model", "разложи по DDD", "ограниченные контексты", "доменная модель", "/ddd-decompose".

Fast path for recording a routine decision as a light ADR (200-400 words, 6-section DDR shape). Use when you just chose one approach over another and want to capture the why before it slips out of memory. NOT for big architecture — that's `adr-architect` agent territory. Triggers: "record decision", "/decision", "запиши решение", "quick adr", "light adr", "ddr"

Disciplined diagnosis loop for hard bugs and performance regressions. Six phases — build feedback loop → reproduce → hypothesise → instrument → fix + regression test → cleanup + post-mortem. Phase 1 (the feedback loop) is the entire skill — everything else is mechanical once you have a fast deterministic pass/fail signal. On verified fix, captures **Evidence via MCP-first path** (`mcp__forgeplan__forgeplan_new` + `forgeplan_link`) with CLI fallback when MCP is not connected. Use when a bug is non-trivial, intermittent, performance-related, or has resisted obvious fixes. Triggers (EN/RU) — "diagnose this", "debug this", "find the root cause", "performance regression", "intermittent bug", "продиагностируй", "найди причину бага", "отладь", "/diagnose".

Methodology: 3-tier pipeline self-healing (AUTO / ADI / USER classification per PRD-032 Sprint D; extended with 4 anomaly kinds in Sprint F PRD-034 + 1 anomaly kind in Sprint M PRD-039 = 9 total). EN: Explicit draft sweep skill for marketplace workspaces. Scans all draft artifacts via mcp__forgeplan__forgeplan_list, classifies each into 9 outcomes (ready_to_activate / incomplete / stale_unreferenced / ambiguous / orphan_link / mistyped_based_on / expired_evidence / phase_mismatch / link_direction_footgun) mapped to 3 resolution tiers (AUTO = silent activate with batch confirmation, ADI = FPF reasoning loop, USER = per-artifact explicit decision). Executes AUTO tier after user batch-confirm; dispatches ADI sub-agents for ambiguous cases; surfaces USER tier per artifact for explicit decisions. Use at end of /forge-cycle or /autorun runs, or manually when forgeplan_health shows accumulating drafts. Closes Anomaly #7 from Sprint A+B+C log. RU: Explicit-скилл для draft sweep маркетплейс воркспейсов. Сканирует все

Structured report formats for Claude — card-based, human-readable, no ASCII frame noise. Use when finishing a multi-step task (≥5 tool calls, ≥3 files modified, multi-task TaskList, or cross-system effects like git push/PR/deploy). Output-only skill, no forgeplan dependency (per PRD-022 Tier B): reads no MCP state, produces plain markdown for the conversation. Triggers on: report, summary, summarise, отчёт, итоги, завершение задачи, build complete, audit complete, deployment complete, decision recorded, incident resolved, migration done. Provides 5 templates (build/audit/decision/incident/migration), anchor conventions (section icons, plain-prose TL;DR, confidence labels), required sections (not-done, reversibility, drift-risks, next-steps), and 5 anti-patterns. Not for simple Q&A or single-file edits.

One-command project bootstrap for the ForgePlan ecosystem. Probes the forgeplan CLI, runs `forgeplan init`, wires `.mcp.json` and `.claude/settings.json`, then chains `/bootstrap` (universal CLAUDE.md template) and `/setup` (docs/agents/ wizard) so a fresh repo is fully wired in one shot. Recommends — but does not install — companion plugins (fpf, laws-of-ux, agents-core, forgeplan-workflow, forgeplan-orchestra). v2.0: Adds optional `--canonize` step that scaffolds the canonical agent layer (project-agent-matrix.yaml + project-config.yaml + Hindsight mental models) for forgeplan-aware projects per PRD-026 Phase 5. Use on a brand-new project, or on an existing project that has none of `.forgeplan/`, `CLAUDE.md`, `docs/agents/`. Triggers (EN/RU) — "fpl init", "init project", "bootstrap forgeplan", "set up everything", "full project setup", "/fpl-init", "поставь всё", "разверни проект с нуля", "инициализируй проект", "fpl init canonize", "setup agent matrix", "v2 bootstrap".

Project-agnostic GitHub Projects v2 integration. Reads per-project config from .forgeplan/state/gh-project.yaml; never hardcodes project numbers. Five operations — init (one-time setup), add-pr (current/given PR → board), link-prd (Standard+ PRD → matching GH issue + board entry with Forgeplan-ID field), sync-status (read forgeplan artifact status, write project board Status field), list (current items on board). **MCP-first read-only per PRD-022 Tier A** — uses `mcp__forgeplan__forgeplan_get` and `mcp__forgeplan__forgeplan_validate` when MCP available (no mutating ops on forgeplan side); CLI fallback (`bash forgeplan get|validate|list`) when MCP not connected. Gracefully degrades when gh CLI scopes are missing or config absent. Triggers (EN/RU) — "gh-project", "add to project", "link prd to issue", "sync project status", "/gh-project", "добавь на доску", "связи с проектом", "обнови статус на доске".

For any artifact ID, report which of the EPIC-001 4-layer pipeline layers (S10 FPF design, S11 BMAD quality gate, S12 OpenSpec structure, S13 Forgeplan automation) are satisfied and which need work. Also checks C4 coverage for ≥3-module architectural decisions. Returns a per-layer score 0-2 + aggregate percentage + concrete action items per gap. Read-only — never auto-fixes, never writes to forgeplan. Use before activating any Standard+ artifact as a final sanity check. Triggers: "methodology check", "check methodology", "4-layer coverage", "layer report", "/methodology-check", "проверь методологии", "layer coverage", "pipeline check", "coverage report", "method check"

One-command migration from dev-toolkit to fpl-skills. Probes installed plugins, scans CLAUDE.md and project docs for /dev-toolkit:* references, offers Mode A (side-by-side, zero risk) or Mode B (clean cut), executes the steps from docs/MIGRATION-DEV-TOOLKIT-TO-FPL-SKILLS.md. **One-time bootstrap, no runtime forgeplan calls** (per PRD-022 Tier B): operates on local plugin manifests + project file paths; the migrated fpl-skills become MCP-first separately at their own runtime. Triggers (EN/RU) — "migrate from dev-toolkit", "switch to fpl-skills", "переедь на fpl-skills", "переехать с dev-toolkit", "перейти на fpl-skills", "/migrate-from-dev-toolkit".

Methodology: Live state renderer (read-only snapshot, no new event system). EN: Renders structured snapshot of running orchestrator state — current sprint, phase, wave, agents in flight (via forgeplan_claims), TaskList progress, forgeplan_health, recent EVID, files modified count, ETA estimate. Read-only: consumes existing data sources (forgeplan MCP claims/health/list, in-session TaskList, .forgeplan/session.yaml if present, git log). Does NOT introduce a new event system. Single snapshot per invocation — user re-invokes for updated view. Use during long /sprint or /autorun runs when you want a "where are we" check without scrolling agent output. RU: Рендерит структурированный snapshot состояния запущенного оркестратора — текущий sprint, phase, wave, агенты в полёте (через forgeplan_claims), прогресс TaskList, forgeplan_health, недавние EVID, count файлов в коммитах, оценка ETA. Read-only: использует существующие data sources (forgeplan MCP claims/health/list, in-session TaskList, .forgeplan/session.yaml есл

Methodology: Project-scoped agent generation via tech-stack detection. EN: Detects project tech stack from package manifests (package.json, Cargo.toml, go.mod, pyproject.toml, pom.xml, mix.exs, Gemfile, composer.json, Package.swift, pubspec.yaml) + secondary signals (tailwind.config.*, next.config.*, nuxt.config.*, docker-compose.yml). Classifies signals into a stack profile, proposes 1–3 project-scoped agents (project:<slug>-pro) baked from agent-template.md with stack-specific patterns. Asks approval per proposal before writing any file. Does NOT auto-create — user always confirms. Writes agent files to .claude/agents/ on approval and suggests project-agent-matrix.yaml entries for dispatch overrides. RU: Определяет технологический стек проекта по package-манифестам + вторичным сигналам. Классифицирует в stack profile, предлагает 1–3 project-scoped агента (project:<slug>-pro) на базе agent-template.md. Спрашивает подтверждения для каждого предложения. Записывает только после согласия пользователя. НЕ создаёт

Interview-driven refinement of a plan, design, or proposal. Walks the user down each branch of the design tree one question at a time, sharpening fuzzy terminology, surfacing contradictions against the existing domain model, stress-testing edge cases with concrete scenarios, and updating CONTEXT.md and ADRs inline as decisions crystallise. Use when a plan is rough, when terms are being used loosely, when the user said "thoughts?" or "does this make sense?", or before kicking off a sprint where ambiguity will cost time. When forgeplan is reachable, **MCP-first** — recommends `mcp__forgeplan__forgeplan_new` for ADRs/NOTEs surfaced during refinement, `mcp__forgeplan__forgeplan_link` to connect to parent PRD, `mcp__forgeplan__forgeplan_validate` after refining existing artifact; CLI fallback (`bash forgeplan ...`) when MCP not connected. Adapted from Matt Pocock's grill-with-docs. Triggers (EN/RU) — "refine this plan", "stress-test this", "review my design", "sharpen the spec", "уточни план", "проверь на прочност