peaks-loop 4.0.0-beta.3 → 4.0.0-beta.5

package/.claude-plugin/marketplace.json

@@ -34,6 +34,11 @@
         "irreversible-action gates"
       ],
       "skills": [
+        {
+          "name": "peaks-solo",
+          "path": "./skills/peaks-solo",
+          "description": "Dispatcher (分诊员) — natural-language front door for the Peaks-Loop skill family. Use when the user describes a task in NL and does not know which peaks-* skill fits. NOT for code/content/doctor/issue-fix-orchestrator/SOP work (use the specific skill)."
+        },
         {
           "name": "peaks-code",
           "path": "./skills/peaks-code",

package/CHANGELOG.md

@@ -2,6 +2,103 @@
 
 ## [Unreleased]
 
+## 4.0.0-beta.5 — 2026-07-08
+
+### Added — peaks-solo dispatcher (分诊员)
+
+- **`peaks-solo` skill** — `skills/peaks-solo/SKILL.md` + 3 references (triage / fallback / sediment). Natural-language front door for the Peaks-Loop skill family. Use when the user describes a task in NL and does not know which peaks-* skill fits. 0 breaking change: 3.x / 4.x `/peaks-code` / `/peaks-content` / `/peaks-doctor` etc. continue to work.
+- **`peaks skill search` CLI** — `src/services/skill/skill-search-service.ts` + `src/cli/commands/skill-search-commands.ts`. Query / tag / domain filters; substring match; structured JSON output. Used by `peaks-solo` to find the right leaf skill. Available as a top-level primitive (not dispatcher-specific).
+- **Sub-skills unchanged** — `peaks-code / peaks-content / peaks-doctor / peaks-issue-fix-orchestrator / peaks-sop / etc.` are NOT modified. peaks-solo sits alongside, not on top.
+
+### Verification (4.0.0-beta.5)
+
+- `peaks skill list` shows `peaks-solo` first
+- `peaks skill search --query "code"` returns `peaks-code` with matchScore > 0
+- `peaks skill search --query "xxxxxxxxxxxxx"` returns `[]` (no error)
+- `pnpm vitest run` — full regression green; `peaks-code / peaks-content / peaks-doctor` tests unchanged
+- `peaks code --help` / `peaks content --help` / `peaks doctor --help` — exit 0, behavior unchanged
+
+## 4.0.0-beta.4 — 2026-07-08
+
+Loop Engineering crystallization is now the product surface. The 4.0.0-beta.3 line shipped the framework; this release ships the post-run crystallization engine, the Darwin-style ratchet, the bundle share / desktop extension surface, and the karpathy-engineered red-line set that locks them together. Every durable-change entry is gated on a real, completed run; every evolution round is gated on an independent-context evaluation.
+
+### Added — Loop Engineering Asset layer (§4.1, AC-1/AC-2/AC-3)
+
+- **`loop_release` table** — `src/services/skillhub/migrations/002-loop-release.sql`. Schema version pinned to `peaks.loop/1` via CHECK constraint. Indexes on `lifecycle_status` and `scenario`. Non-breaking: 4.x `bee_release` rows continue to read.
+- **`loop_bee_relation` table** — `src/services/skillhub/migrations/003-loop-bee-relation.sql`. Schema version `peaks.loop-bee-relation/1`. Roles: `main` / `supporting` / `candidate` / `retired`. Partial UNIQUE index `WHERE role='main'` enforces at most one main bee per loop at the storage layer (defense in depth; the service layer also enforces it).
+- **Schema for the four-loop + dual-asset model** — Zod schemas in `src/services/loop/loop-release-types.ts` and `src/services/loop/loop-bee-relation-types.ts` mirror the §4.1 / §4.6 row shapes and refuse to parse rows whose `schema_version` is not the fixed literal.
+
+### Added — Post-run crystallization flow (§5, AC-4 / AC-5 / AC-6 / AC-7)
+
+- **`crystallization_event` table** — `src/services/skillhub/migrations/006-crystallization-event.sql`. Schema version `peaks.crystallization/1`. Carries the 4-section evidence brief inline; FKs to optional created/updated loop_release / bee_release.
+- **`CrystallizationService`** — `src/services/crystallization/crystallization-service.ts`. Pre-run gate: `task_status='completed'` AND `gates_passed=true` AND `evidence_collected=true` (Zod literal enforcement + service-layer re-assertion). Atomic single-transaction write of loop_release + bee_release header + bee_manifest + loop_bee_relation(main) + crystallization_event(brief inline).
+- **`peaks asset crystallize`** CLI — `src/cli/commands/asset-commands.ts`. Required options: `--brief-what-happened`, `--brief-why-it-matters`, `--brief-what-learned`, `--brief-what-action`. A partial brief is rejected with `MISSING_BRIEF_SECTION` and exit 1 (RL-7).
+- **`peaks asset dispose` / `peaks asset status`** — cross-asset dispose (trace_only / retain / destroy) and lifecycle status dashboard.
+
+### Added — Darwin-style ratchet (§6, AC-8 / AC-9 / AC-10 / AC-11)
+
+- **`evolution_evaluation` table** — `src/services/skillhub/migrations/005-evolution-evaluation.sql`. Schema version `peaks.evolution/1`. Carries `target_kind`, `target_release_id`, `optimization_dimensions_json` (length-1 enforced at service layer), `target_count=1` (CHECK), `author_id`, `evaluator_id`, `skeptic_id` (three independent agents), `verdict` ∈ {`keep`, `revert`, `needs-user-decision`}, `user_confirmation_pointer`, `brief_pointer`, and the four-section brief projection.
+- **`EvolutionService`** — `src/services/evolution/evolution-service.ts`. Hard rules: single object (AC-8) + single optimization dimension (AC-8) + author ≠ evaluator ≠ skeptic (AC-10/AC-12/AC-14) + score_delta >= score_delta_min for `keep` (AC-11, default 1.0) + user_confirmation_pointer required for `keep` (AC-15).
+- **`peaks evolution propose / evaluate / revert / mark-keep / status`** CLI — `src/cli/commands/evolution-commands.ts`. Each error path emits a stable wire-format code (`EVOLUTION_MULTI_OBJECT` / `EVOLUTION_MULTI_DIMENSION` / `EVOLUTION_SELF_SCORE` / `EVOLUTION_DELTA_BELOW_THRESHOLD` / `EVOLUTION_MISSING_USER_CONFIRMATION`).
+- **Independent-evaluator runner** — `src/services/evolution/independent-evaluator-runner.ts`. Frozen `EvaluationPackage` containing only `target_kind / target_release_id / optimization_dimension / before_snapshot / after_snapshot / diff / rubric / red_lines / source_traces` — no author session, no author reasoning, no recommendation framing (AC-12 / AC-13).
+- **Regression-skeptic runner** — `src/services/evolution/regression-skeptic-runner.ts`. A SEPARATE agent that emits `driftRisks / overfitRisks / safetyRegressionRisks` plus an optional `blocker` (which forces `verdict='revert'`).
+
+### Added — Karpathy-engineered red-line set (AC-21 / AC-22 / AC-23)
+
+- **`.peaks/standards/loop-engineering-guidelines.md`** — the single source of truth for the 10 red lines RL-0..RL-9, each in the four-section karpathy form (Failure modes / Rewrite / Self-check / Out-of-scope). Co-equal karpathy × darwin layers (RL-0).
+- **`peaks standards lint --category loop-engineering`** — `src/services/standards/loop-engineering-lint.ts`. Parses the guideline file and asserts every red line has all four sections, fails closed otherwise.
+- **`peaks skill lint --category loop-engineering-readiness --path <skill-dir>`** — `src/services/standards/loop-engineering-readiness-lint.ts` + `src/cli/commands/skill-loop-engineering-readiness-commands.ts`. Asserts a peaks-* SKILL.md (a) references `.peaks/standards/loop-engineering-guidelines.md`, (b) does not introduce a CLI verb the user is meant to type, (c) does not introduce a JSON / manifest hand-authoring surface. Alias verb: `peaks skill ready --category loop-engineering-readiness --path <skill-dir>`.
+- **Unit guard** — `tests/unit/standards/loop-engineering-guidelines.test.ts` enforces the four-section shape for every red line in the file.
+- **peaks-code domain boundary (RL-8)** — `skills/peaks-code/SKILL.md` self-declares as the code-domain long-task loop engineering orchestrator and is NOT a general-purpose orchestrator. Cross-domain peaks-* skills (peaks-content, peaks-issue-fix-orchestrator) import the shared guideline file and pass the readiness lint.
+
+### Added — Bundle share + desktop extension surface (§7A, AC-24 / AC-25 / AC-26)
+
+- **Share / desktop fields** — `src/services/skillhub/migrations/004-loop-bee-extension.sql`. Adds `shareable / share_excluded_paths / desktop_visible / export_bundle_format` to `loop_release`, and `shareable / desktop_visible` to `bee_release`. `export_bundle_format` is pinned to `peaks.bundle/1` via CHECK constraint. Non-breaking: every new column has a DEFAULT.
+- **Bundle writer** — `src/services/share/bundle-writer.ts`. Writes a `peaks.bundle/1` tar.gz with `manifest.json`, `relations.json`, `evidence_briefs/*.json`, content-addressed `blobs/<sha256>`, and `EVALUATION_REQUIRED.md`. Hard block: `shareable=false` throws `BundleNotShareableError(SHARE_BUNDLE_ERROR_CODES.NOT_SHAREABLE)` BEFORE any tarball work, for both loop and bee kinds.
+- **Bundle reader** — `src/services/share/bundle-reader.ts`. Hard blocks: `format_version_major != 1` is rejected (AC-25); `schema_versions` mismatch is rejected; the imported release ALWAYS lands as `candidate` (AC-25 / §10 RL-9). The bundle carries an `EVALUATION_REQUIRED.md` marker telling the receiver to run an independent evaluation before any durable change.
+- **`peaks loop export` / `peaks loop import`** — `src/cli/commands/loop-commands.ts`. Existing 14.x `peaks loop *` subcommands are preserved; M7 only ADDS export / import.
+- **`peaks bee export` / `peaks bee import`** — `src/cli/commands/share-commands.ts`. Same hard-block semantics.
+- **Receiver-side integration test** — `tests/integration/share-bundle-roundtrip.test.ts` asserts the round-trip lands as `candidate` (AC-25), that `shareable=false` blocks the export at the CLI layer, and that without an `evolution_evaluation` row the receiver has no path to `stable` (AC-26).
+
+### Added — SkillHub expansion (5 new tables / non-breaking)
+
+The 6+ relation table layout is preserved. 5 new tables added under `.peaks/_runtime/<sessionId>/`:
+- `loop_release` (002)
+- `loop_bee_relation` (003)
+- loop↔bee share / desktop extension columns (004)
+- `evolution_evaluation` (005)
+- `crystallization_event` (006)
+
+Migration is non-breaking: every new column has a DEFAULT and the new tables sit alongside the existing 4.x schema.
+
+### Added — New CLI verbs (spec §7.4)
+
+- `peaks loop init / list / show / search / recent / export / import`
+- `peaks asset crystallize / dispose / status`
+- `peaks evolution propose / evaluate / revert / mark-keep / status`
+- `peaks skill lint --category loop-engineering-readiness --path <skill-dir>`
+- `peaks skill ready --category loop-engineering-readiness --path <skill-dir>` (alias)
+
+### Changed
+
+- **`peaks workflow *`** is reframed as the execution trace surface (per spec §7.6 demotion); it remains functional. The user-facing verb is "replay this run", not "create a new asset".
+- **peaks-maker skill** is narratively repositioned to "Loop crystallizer + Bee creator + Evolution gatekeeper" (per spec §7.5); only the SKILL.md and memory references are re-narrated, the `id` is preserved.
+
+### Verification (4.0.0-beta.4)
+
+- `tsc --noEmit` zero errors
+- `peaks standards lint --category loop-engineering` parses all 10 red lines + 4 sections each (lint passes)
+- `pnpm vitest run tests/unit/{standards,crystallization,evolution,share}/` — 13 files / 154 unit tests pass
+- `pnpm vitest run tests/integration/{share-bundle-roundtrip,dogfood-loop-engineering-crystallization,asset-crystallize-cli,evolution-cli,skill-loop-engineering-readiness-cli}.test.ts` — 5 files / 26 integration tests pass
+- M8 dogfood (`tests/integration/dogfood-loop-engineering-crystallization.test.ts`) crystallizes the M0..M7 work into a real loop + bee + relation + event in 644ms (defense in depth: the release crystallizes itself)
+
+### Migration
+
+- 4.x `bee_release` rows continue to read. No data migration required.
+- The 5 new tables are added via non-destructive `CREATE TABLE IF NOT EXISTS` migrations.
+- The `bundle-writer / bundle-reader` pair is additive; existing `peaks skill sediment export / import` remains as an alias for one release cycle, then deprecates.
+- The peaks-code SKILL.md is unchanged from 4.0.0-beta.3 — no `peaks-solo` migration needed for users already on beta.2 or later.
+
 ## 4.0.0-beta.2 — 2026-07-07
 
 ### Renamed

package/README-en.md

@@ -33,6 +33,10 @@ The product is governed by a four-layer asset model and a karpathy × darwin dua
 
 Summon them, they handle the work. Stand them down when it's done. **No skipped steps. No half-finished hand-offs.** No new AI CLI to learn — it sits on top of Claude Code, Codex, or Copilot you already run.
 
+### peaks-solo is the dispatcher (new in 4.0.0-beta.5)
+
+If you **don't know which peaks-* skill to use**, just type `/peaks-solo` and describe your task in natural language. It triages for you: dispatches to a matching leaf, or self-plans with deep-search / WebSearch / Bash / Edit markdown if no leaf fits, then asks whether to sediment the result. `/peaks-code` / `/peaks-content` / `/peaks-doctor` etc. continue to work — **zero breaking changes**.
+
 **Every gate has a hard exit.** Audit fails = stop. QA fails = stop. Any gate fails = full stop. You don't chase it, you don't patch it. You decide, it runs the next gate.
 
 #### Gates aren't decoration — 5,439 tests passed · 19 skipped · 0 failed

package/README.md

@@ -33,6 +33,10 @@
 
 召之即来,事完收队,**不跳步,不半截扔给你**。它不是新发明一个 AI CLI,而是架在你已有的 Claude Code / Codex / Copilot 之上,把它们都调成同一支团队,只对你这一个开发者的口味。
 
+### peaks-solo 是分诊员(新增,4.0.0-beta.5)
+
+如果**你不知道该用哪个 peaks-* 技能**,直接用 `/peaks-solo` 描述你的诉求就行。它会替你分诊:有合适的 leaf 就透明转交,没合适的就自己规划 + 跑(deep-search / WebSearch / Bash / Edit markdown),跑完回头问你要不要沉淀。`/peaks-code` / `/peaks-content` / `/peaks-doctor` 等老入口照常可用,**0 breaking**。
+
 **每一步都有强退出条件**:审计不通过就停,QA 没过就停,任何一道门失败 = 全流程停。你不必追着它催、也不必替它补,它跑完一道再交你拍板,拍完再走下一道。
 
 #### 门禁不是装饰 —— 5439 tests passed · 19 skipped · 0 failed

package/dist/src/cli/commands/core/skill-command.js

@@ -9,6 +9,15 @@ import { generateProjectContext } from '../../../services/memory/project-context
 import { getSessionId, setSessionMeta } from '../../../services/session/session-manager.js';
 import { fail, ok } from '../../../shared/result.js';
 import { addJsonOption, getErrorMessage, printResult } from '../../cli-helpers.js';
+// Slice S0 (4.0.0-beta.5 peaks-solo dispatcher release):
+// `peaks skill search` is the CLI primitive that feeds the
+// peaks-solo dispatcher (S1). Adding a single import + register call
+// here keeps the change surgical and leaves all existing skill
+// subcommands (list / doctor / sync / runbook / presence / heartbeat)
+// untouched. See
+// docs/superpowers/specs/2026-07-08-peaks-solo-dispatcher-design.md §3.2
+// and the S0 plan under docs/superpowers/plans/.
+import { registerSkillSearchCommand } from '../skill-search-commands.js';
 export function registerSkillCommand(program, io) {
     const skill = program.command('skill').description('Manage Peaks skills');
     addJsonOption(skill.command('list').description('List skills derived from skills/*/SKILL.md')).action(async (options) => {
@@ -18,6 +27,15 @@ export function registerSkillCommand(program, io) {
         }
         else {
             const sorted = [...skills].sort((a, b) => {
+                // Slice S0 (4.0.0-beta.5): peaks-solo is the dispatcher (front
+                // door) — list it FIRST so users discover the dispatcher before
+                // any specific leaf. Followed by peaks-sop (current default
+                // runbook showcase) and peaks-code (canonical code-domain
+                // orchestrator); everything else falls back to alphabetical.
+                if (a.name === 'peaks-solo')
+                    return -1;
+                if (b.name === 'peaks-solo')
+                    return 1;
                 if (a.name === 'peaks-sop')
                     return -1;
                 if (b.name === 'peaks-sop')
@@ -272,4 +290,8 @@ export function registerSkillCommand(program, io) {
         const result = detectPresenceMarker({ project: projectRoot, latestAssistantMessage: message });
         printResult(io, ok('skill.detect-marker-loss', result), options.json);
     });
+    // Slice S0 — register `peaks skill search`. Sibling subcommand to
+    // list / runbook / presence; preserves the existing surface
+    // (HC-10 — 老入口保留).
+    registerSkillSearchCommand(program, io);
 }

package/dist/src/cli/commands/skill-search-commands.d.ts

@@ -0,0 +1,25 @@
+/**
+ * peaks skill search — CLI primitive for S0 of the 4.0.0-beta.5
+ * peaks-solo dispatcher release.
+ *
+ * Spec: docs/superpowers/specs/2026-07-08-peaks-solo-dispatcher-design.md §3.2
+ * Plan: docs/superpowers/plans/2026-07-08-peaks-solo-dispatcher/s0-skill-search-cli.md
+ *
+ * Wired into the existing `peaks skill` group by
+ * `src/cli/commands/core/skill-command.ts` (one import + one call).
+ * The command ALWAYS emits a JSON array on stdout, even on the
+ * no-match case (which exits 0 with `[]` per plan §"Match rules (v1)").
+ *
+ * Exit codes (locked in plan §"API Contract"):
+ *   0 — success (may be empty array)
+ *   1 — invalid args (Zod validation failure)
+ *   2 — unexpected error (skill pool read fail, etc.)
+ */
+import type { Command } from 'commander';
+import type { ProgramIO } from '../cli-helpers.js';
+/**
+ * Validate the raw CLI flags, call the service, and emit a JSON
+ * envelope on stdout. Errors flow through stderr + non-zero exit
+ * codes per the plan §"API Contract" table.
+ */
+export declare function registerSkillSearchCommand(program: Command, io: ProgramIO): void;

package/dist/src/cli/commands/skill-search-commands.js

@@ -0,0 +1,68 @@
+import { searchSkills, SkillSearchInputSchema } from '../../services/skill/skill-search-service.js';
+import { getErrorMessage } from '../../shared/result.js';
+/**
+ * Validate the raw CLI flags, call the service, and emit a JSON
+ * envelope on stdout. Errors flow through stderr + non-zero exit
+ * codes per the plan §"API Contract" table.
+ */
+export function registerSkillSearchCommand(program, io) {
+    // The parent `skill` command is created by registerSkillCommand in
+    // core/skill-command.ts. To keep the wiring self-contained we
+    // resolve the parent by name; this also lets tests register the
+    // subcommand in isolation.
+    const skill = program.commands.find((c) => c.name() === 'skill');
+    if (skill === undefined) {
+        // Defensive: the parent must be registered before us. Surface the
+        // error visibly — the dispatcher cannot triage without `peaks skill`
+        // existing, so the LLM-side caller needs to know.
+        throw new Error('peaks skill search: parent `skill` command not found. ' +
+            'Ensure registerSkillCommand is called before registerSkillSearchCommand.');
+    }
+    skill
+        .command('search')
+        .description('Search the in-tree skill pool by --query (substring on description + triggers) ' +
+        'and/or --tag (exact match on metadata.tags) and/or --domain (exact match). ' +
+        'Outputs a JSON array (always, even on no-match) to stdout. ' +
+        'See docs/superpowers/specs/2026-07-08-peaks-solo-dispatcher-design.md §3.2.')
+        .option('-q, --query <text>', 'case-insensitive substring matched against skill description + triggers')
+        .option('-t, --tag <tag>', 'exact match on metadata.tags (e.g. "orchestrator", "loop-engineering")')
+        .option('-d, --domain <domain>', 'exact match on metadata.domain (locked enum: code | content | doctor | research | triage | sop | audit | final-review | resume | status | test | ide | slice-decompose | issue-fix-orchestrator | perf-audit | security-audit | reviewer)')
+        .option('-l, --limit <n>', 'max results returned (1..100, default 20)', (v) => Number.parseInt(v, 10))
+        .action(async (options) => {
+        // 1) Coerce + validate input via the service's Zod schema.
+        //    The schema's refine rejects empty input — that is the
+        //    "all-empty → error" contract the spec mandates.
+        const raw = {};
+        if (options.query !== undefined)
+            raw['query'] = options.query;
+        if (options.tag !== undefined)
+            raw['tag'] = options.tag;
+        if (options.domain !== undefined)
+            raw['domain'] = options.domain;
+        if (options.limit !== undefined)
+            raw['limit'] = options.limit;
+        const parsed = SkillSearchInputSchema.safeParse(raw);
+        if (!parsed.success) {
+            // Exit code 1 = invalid args. Emit a human-readable message
+            // on stderr so the LLM-side caller can act on it; do NOT
+            // pollute stdout (which downstream tools parse as JSON).
+            for (const issue of parsed.error.issues) {
+                io.stderr(`${issue.path.join('.') || '<root>'}: ${issue.message}`);
+            }
+            process.exitCode = 1;
+            return;
+        }
+        // 2) Call the service. Any unexpected error (skill pool read
+        //    fail, etc.) maps to exit code 2.
+        try {
+            const results = await searchSkills(parsed.data);
+            // Always emit a JSON array on stdout. The plan locks this
+            // contract: even no-match returns `[]`, never `null`.
+            io.stdout(JSON.stringify(results));
+        }
+        catch (error) {
+            io.stderr(`skill.search failed: ${getErrorMessage(error)}`);
+            process.exitCode = 2;
+        }
+    });
+}

package/dist/src/services/context/context-schema.d.ts

@@ -223,39 +223,39 @@ export declare const ContextJsonSchema: z.ZodObject<{
             tags: z.ZodArray<z.ZodString, "many">;
         }, "strip", z.ZodTypeAny, {
             kind: "git" | "code" | "doc" | "memory";
+            tags: string[];
             id: string;
             blastRadius: string[];
             conflictScore: number;
             timeDecayScore: number;
-            tags: string[];
             version?: string | undefined;
         }, {
             kind: "git" | "code" | "doc" | "memory";
+            tags: string[];
             id: string;
             blastRadius: string[];
             conflictScore: number;
             timeDecayScore: number;
-            tags: string[];
             version?: string | undefined;
         }>, "many">;
     }, "strip", z.ZodTypeAny, {
         metadata: {
             kind: "git" | "code" | "doc" | "memory";
+            tags: string[];
             id: string;
             blastRadius: string[];
             conflictScore: number;
             timeDecayScore: number;
-            tags: string[];
             version?: string | undefined;
         }[];
     }, {
         metadata: {
             kind: "git" | "code" | "doc" | "memory";
+            tags: string[];
             id: string;
             blastRadius: string[];
             conflictScore: number;
             timeDecayScore: number;
-            tags: string[];
             version?: string | undefined;
         }[];
     }>;
@@ -330,11 +330,11 @@ export declare const ContextJsonSchema: z.ZodObject<{
     tokenizer: {
         metadata: {
             kind: "git" | "code" | "doc" | "memory";
+            tags: string[];
             id: string;
             blastRadius: string[];
             conflictScore: number;
             timeDecayScore: number;
-            tags: string[];
             version?: string | undefined;
         }[];
     };
@@ -397,11 +397,11 @@ export declare const ContextJsonSchema: z.ZodObject<{
     tokenizer: {
         metadata: {
             kind: "git" | "code" | "doc" | "memory";
+            tags: string[];
             id: string;
             blastRadius: string[];
             conflictScore: number;
             timeDecayScore: number;
-            tags: string[];
             version?: string | undefined;
         }[];
     };

package/dist/src/services/skill/skill-search-service.d.ts

@@ -0,0 +1,88 @@
+/**
+ * peaks skill search — service-layer primitive for S0 of the
+ * 4.0.0-beta.5 peaks-solo dispatcher release.
+ *
+ * Spec: docs/superpowers/specs/2026-07-08-peaks-solo-dispatcher-design.md §3.2
+ * Plan: docs/superpowers/plans/2026-07-08-peaks-solo-dispatcher/s0-skill-search-cli.md
+ *
+ * This service is the single source of truth for skill pool retrieval
+ * and is reused by the `peaks skill search` CLI. It is intentionally
+ * pure (synchronous frontmatter parse + substring scoring) so the
+ * dispatcher (`peaks-solo`, S1) can call it without paying for any
+ * async I/O during the triage decision flow.
+ *
+ * v1 scope (locked, see plan §"Match rules (v1)"):
+ *   - --query: case-insensitive substring match on description + triggers
+ *   - --tag:   case-insensitive exact match on `metadata.tags`
+ *   - --domain: exact match on `metadata.domain` (locked enum)
+ *   - All three compose with AND; at least one filter is required
+ *   - No match → `[]` (not error, not null)
+ *   - --limit truncates to the top N by matchScore desc
+ *
+ * Out of v1 scope (deferred, see plan §"Goal"): fuzzy / FTS5 / semantic
+ * matching, ML-based ranking, pagination, i18n, caching.
+ */
+import { z } from 'zod';
+import { skillsDir } from '../../shared/paths.js';
+export declare const SkillSearchInputSchema: z.ZodEffects<z.ZodObject<{
+    query: z.ZodOptional<z.ZodString>;
+    tag: z.ZodOptional<z.ZodString>;
+    domain: z.ZodOptional<z.ZodEnum<["code", "content", "doctor", "research", "triage", "sop", "audit", "final-review", "resume", "status", "test", "ide", "slice-decompose", "issue-fix-orchestrator", "perf-audit", "security-audit", "reviewer"]>>;
+    limit: z.ZodDefault<z.ZodOptional<z.ZodNumber>>;
+}, "strip", z.ZodTypeAny, {
+    limit: number;
+    query?: string | undefined;
+    tag?: string | undefined;
+    domain?: "code" | "status" | "content" | "test" | "doctor" | "resume" | "ide" | "research" | "triage" | "sop" | "audit" | "final-review" | "slice-decompose" | "issue-fix-orchestrator" | "perf-audit" | "security-audit" | "reviewer" | undefined;
+}, {
+    limit?: number | undefined;
+    query?: string | undefined;
+    tag?: string | undefined;
+    domain?: "code" | "status" | "content" | "test" | "doctor" | "resume" | "ide" | "research" | "triage" | "sop" | "audit" | "final-review" | "slice-decompose" | "issue-fix-orchestrator" | "perf-audit" | "security-audit" | "reviewer" | undefined;
+}>, {
+    limit: number;
+    query?: string | undefined;
+    tag?: string | undefined;
+    domain?: "code" | "status" | "content" | "test" | "doctor" | "resume" | "ide" | "research" | "triage" | "sop" | "audit" | "final-review" | "slice-decompose" | "issue-fix-orchestrator" | "perf-audit" | "security-audit" | "reviewer" | undefined;
+}, {
+    limit?: number | undefined;
+    query?: string | undefined;
+    tag?: string | undefined;
+    domain?: "code" | "status" | "content" | "test" | "doctor" | "resume" | "ide" | "research" | "triage" | "sop" | "audit" | "final-review" | "slice-decompose" | "issue-fix-orchestrator" | "perf-audit" | "security-audit" | "reviewer" | undefined;
+}>;
+export type SkillSearchInput = z.input<typeof SkillSearchInputSchema>;
+export declare const SkillSearchResultSchema: z.ZodObject<{
+    name: z.ZodString;
+    description: z.ZodString;
+    triggers: z.ZodArray<z.ZodString, "many">;
+    tags: z.ZodArray<z.ZodString, "many">;
+    domain: z.ZodString;
+    matchScore: z.ZodNumber;
+}, "strip", z.ZodTypeAny, {
+    name: string;
+    description: string;
+    domain: string;
+    triggers: string[];
+    tags: string[];
+    matchScore: number;
+}, {
+    name: string;
+    description: string;
+    domain: string;
+    triggers: string[];
+    tags: string[];
+    matchScore: number;
+}>;
+export type SkillSearchResult = z.infer<typeof SkillSearchResultSchema>;
+/**
+ * Search the in-tree skill pool. Validates `input` against
+ * `SkillSearchInputSchema`; the schema's refine rejects empty input
+ * with a ZodError, which is the contract test U-7 / I-3 rely on.
+ *
+ * @param input  One of `query | tag | domain` is required.
+ *               All provided filters compose with AND.
+ * @returns      Matching skills sorted by matchScore desc, name asc.
+ *               Empty array (not error, not null) when nothing matches.
+ */
+export declare function searchSkills(rawInput: SkillSearchInput): Promise<SkillSearchResult[]>;
+export { skillsDir as __skillsDirForTests };

package/dist/src/services/skill/skill-search-service.js

@@ -0,0 +1,228 @@
+/**
+ * peaks skill search — service-layer primitive for S0 of the
+ * 4.0.0-beta.5 peaks-solo dispatcher release.
+ *
+ * Spec: docs/superpowers/specs/2026-07-08-peaks-solo-dispatcher-design.md §3.2
+ * Plan: docs/superpowers/plans/2026-07-08-peaks-solo-dispatcher/s0-skill-search-cli.md
+ *
+ * This service is the single source of truth for skill pool retrieval
+ * and is reused by the `peaks skill search` CLI. It is intentionally
+ * pure (synchronous frontmatter parse + substring scoring) so the
+ * dispatcher (`peaks-solo`, S1) can call it without paying for any
+ * async I/O during the triage decision flow.
+ *
+ * v1 scope (locked, see plan §"Match rules (v1)"):
+ *   - --query: case-insensitive substring match on description + triggers
+ *   - --tag:   case-insensitive exact match on `metadata.tags`
+ *   - --domain: exact match on `metadata.domain` (locked enum)
+ *   - All three compose with AND; at least one filter is required
+ *   - No match → `[]` (not error, not null)
+ *   - --limit truncates to the top N by matchScore desc
+ *
+ * Out of v1 scope (deferred, see plan §"Goal"): fuzzy / FTS5 / semantic
+ * matching, ML-based ranking, pagination, i18n, caching.
+ */
+import { z } from 'zod';
+import { listSkills } from '../skills/skill-registry.js';
+import { parseFrontmatter } from '../../shared/frontmatter.js';
+import { readText } from '../../shared/fs.js';
+import { skillsDir } from '../../shared/paths.js';
+const SKILL_DOMAINS = [
+    'code',
+    'content',
+    'doctor',
+    'research',
+    'triage',
+    'sop',
+    'audit',
+    'final-review',
+    'resume',
+    'status',
+    'test',
+    'ide',
+    'slice-decompose',
+    'issue-fix-orchestrator',
+    'perf-audit',
+    'security-audit',
+    'reviewer'
+];
+export const SkillSearchInputSchema = z
+    .object({
+    query: z.string().min(1).max(500).optional(),
+    tag: z.string().min(1).max(100).optional(),
+    domain: z.enum(SKILL_DOMAINS).optional(),
+    limit: z.number().int().min(1).max(100).optional().default(20)
+})
+    .refine((v) => v.query !== undefined || v.tag !== undefined || v.domain !== undefined, { message: 'At least one of --query / --tag / --domain is required' });
+export const SkillSearchResultSchema = z.object({
+    name: z.string(),
+    description: z.string(),
+    triggers: z.array(z.string()),
+    tags: z.array(z.string()),
+    domain: z.string(),
+    matchScore: z.number().min(0).max(1)
+});
+/**
+ * Extract trigger phrases from a skill description. The convention is
+ * `Triggers on \`/x\`, "y", "z".` at the end of the description
+ * (or anywhere within). v1 parses quoted/back-tick tokens after the
+ * "Triggers on" marker. When the marker is missing, the trigger set
+ * is empty (description-only matching still works).
+ */
+function extractTriggers(description) {
+    const lower = description.toLowerCase();
+    const marker = 'triggers on';
+    const idx = lower.indexOf(marker);
+    if (idx === -1)
+        return [];
+    const after = description.slice(idx + marker.length);
+    // Stop at the first period that ends the trigger clause.
+    // Heuristic: take everything up to the next `.` that is followed by
+    // whitespace + capital letter, or end-of-string.
+    const stopMatch = /\.\s+[A-Z]|\.\s*$/.exec(after);
+    const clause = stopMatch ? after.slice(0, stopMatch.index + 1) : after;
+    // Extract every quoted/back-tick token.
+    const tokenRegex = /[`"']([^`"']+)[`"']/g;
+    const out = [];
+    let m;
+    while ((m = tokenRegex.exec(clause)) !== null) {
+        if (m[1] !== undefined)
+            out.push(m[1]);
+    }
+    return out;
+}
+/**
+ * Parse metadata tags / domain from the raw SKILL.md frontmatter.
+ * `parseFrontmatter()` flattens nested YAML mapping into dotted keys,
+ * so `metadata.tags: a, b` becomes a single string we split on `, `.
+ * The existing `skill-registry` already calls `parseFrontmatter` but
+ * does not surface the metadata block, so we re-parse the file.
+ *
+ * R8 mitigation: if the file is missing, malformed, or unreadable we
+ * return empty tags + domain so the search itself can still proceed
+ * with description-only matching. The malformed-file warning is
+ * already emitted by the registry layer (`loadSkillRegistry`'s
+ * `failures[]` channel); this function is best-effort metadata.
+ */
+async function parseMetadata(skillPath) {
+    try {
+        const raw = await readText(skillPath);
+        const fm = parseFrontmatter(raw);
+        const tagsRaw = fm['metadata.tags'] ?? '';
+        const tags = tagsRaw
+            .split(',')
+            .map((t) => t.trim())
+            .filter((t) => t.length > 0);
+        const domain = (fm['metadata.domain'] ?? '').trim();
+        return { tags, domain };
+    }
+    catch {
+        return { tags: [], domain: '' };
+    }
+}
+async function loadEnrichedSkills() {
+    const skills = await listSkills();
+    return Promise.all(skills.map(async (s) => {
+        const meta = await parseMetadata(s.skillPath);
+        return {
+            name: s.name,
+            description: s.description,
+            triggers: extractTriggers(s.description),
+            tags: meta.tags,
+            domain: meta.domain,
+            skillPath: s.skillPath
+        };
+    }));
+}
+/**
+ * Substring match score for a single (skill, query) pair.
+ *
+ * matchScore = 0.5 * (query_hits_in_description / description_length_words)
+ *            + 0.5 * (query_hits_in_triggers / triggers_count)
+ *            clamped to [0, 1]
+ *
+ * "hits" = number of occurrences of the query substring (lowercased)
+ * in the lowercased target. Length is normalized as word count (so a
+ * 1-word description that matches returns the full 0.5 contribution).
+ */
+function scoreQuery(query, description, triggers) {
+    const q = query.toLowerCase();
+    const desc = description.toLowerCase();
+    const descHits = countOccurrences(desc, q);
+    const descWords = Math.max(1, desc.split(/\s+/).filter((w) => w.length > 0).length);
+    const descContribution = Math.min(1, descHits / descWords);
+    const triggersJoined = triggers.map((t) => t.toLowerCase()).join(' | ');
+    const trigHits = countOccurrences(triggersJoined, q);
+    const trigCount = Math.max(1, triggers.length);
+    const trigContribution = Math.min(1, trigHits / trigCount);
+    const raw = 0.5 * descContribution + 0.5 * trigContribution;
+    // Clamp to [0, 1] and round to 4 decimals to keep envelope tidy.
+    return Math.round(Math.max(0, Math.min(1, raw)) * 10_000) / 10_000;
+}
+function countOccurrences(haystack, needle) {
+    if (needle.length === 0)
+        return 0;
+    let count = 0;
+    let pos = 0;
+    while (pos <= haystack.length) {
+        const next = haystack.indexOf(needle, pos);
+        if (next === -1)
+            break;
+        count += 1;
+        pos = next + needle.length;
+    }
+    return count;
+}
+/**
+ * Search the in-tree skill pool. Validates `input` against
+ * `SkillSearchInputSchema`; the schema's refine rejects empty input
+ * with a ZodError, which is the contract test U-7 / I-3 rely on.
+ *
+ * @param input  One of `query | tag | domain` is required.
+ *               All provided filters compose with AND.
+ * @returns      Matching skills sorted by matchScore desc, name asc.
+ *               Empty array (not error, not null) when nothing matches.
+ */
+export async function searchSkills(rawInput) {
+    const input = SkillSearchInputSchema.parse(rawInput);
+    const skills = await loadEnrichedSkills();
+    const filtered = skills.filter((skill) => {
+        if (input.tag !== undefined) {
+            const wanted = input.tag.toLowerCase();
+            const has = skill.tags.some((t) => t.toLowerCase() === wanted);
+            if (!has)
+                return false;
+        }
+        if (input.domain !== undefined) {
+            if (skill.domain.toLowerCase() !== input.domain.toLowerCase())
+                return false;
+        }
+        if (input.query !== undefined) {
+            const q = input.query.toLowerCase();
+            const inDesc = skill.description.toLowerCase().includes(q);
+            const inTrig = skill.triggers.some((t) => t.toLowerCase().includes(q));
+            if (!inDesc && !inTrig)
+                return false;
+        }
+        return true;
+    });
+    const scored = filtered.map((skill) => ({
+        name: skill.name,
+        description: skill.description,
+        triggers: skill.triggers,
+        tags: skill.tags,
+        domain: skill.domain,
+        matchScore: input.query !== undefined
+            ? scoreQuery(input.query, skill.description, skill.triggers)
+            : 1
+    }));
+    scored.sort((a, b) => {
+        if (b.matchScore !== a.matchScore)
+            return b.matchScore - a.matchScore;
+        return a.name.localeCompare(b.name);
+    });
+    return scored.slice(0, input.limit);
+}
+// Re-export `skillsDir` only for tests that want to assert on-disk I/O
+// behavior end-to-end. Not part of the public CLI surface.
+export { skillsDir as __skillsDirForTests };

package/dist/src/shared/version.d.ts

@@ -1 +1 @@
-export declare const CLI_VERSION = "4.0.0-beta.3";
+export declare const CLI_VERSION = "4.0.0-beta.5";

package/dist/src/shared/version.js

@@ -1 +1 @@
-export const CLI_VERSION = "4.0.0-beta.3";
+export const CLI_VERSION = "4.0.0-beta.5";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "peaks-loop",
-  "version": "4.0.0-beta.3",
+  "version": "4.0.0-beta.5",
   "description": "Loop Engineering CLI — workflow primitive / loop guards / evaluators / slice orchestration",
   "author": "SquabbyZ",
   "license": "MIT",

package/skills/peaks-solo/SKILL.md

@@ -0,0 +1,197 @@
+---
+name: peaks-solo
+description: |
+  Dispatcher (分诊员) for the Peaks-Loop skill family. Use when the user describes a task in natural language and does NOT know which peaks-* skill fits. peaks-solo reads the live skill pool via `peaks skill search`, dispatches to a matching leaf (peaks-code / peaks-content / peaks-doctor / peaks-issue-fix-orchestrator / etc.) or falls back to self-planned execution (deep-search / WebSearch / Bash / Edit markdown) if no leaf matches, and then asks the user whether to sediment the result.
+
+  Triggers: 自然语言描述诉求且无明确 peaks-* skill 选择 / "帮我处理这个" / "我不知道该用哪个" / "随便都行".
+
+  NOT for: code-specific work (use /peaks-code) / content-specific work (use /peaks-content) / project health check (use /peaks-doctor) / issue sweep (use /peaks-issue-fix-orchestrator) / SOP authoring (use /peaks-sop).
+metadata:
+  type: dispatcher
+  domain: triage
+  visibility: public
+  supersedes: peaks-solo (eradicated 2026-07-05, revived 2026-07-08)
+  red_lines: [RL-1, RL-8, HC-7, HC-8, HC-9, HC-10, HC-11]
+---
+
+# peaks-solo — Dispatcher (分诊员)
+
+> peaks-solo 是 Peaks-Loop 的 dispatcher,不是 orchestrator。读取技能池 → 分诊 → 转交 leaf 或自规划 → 回头问沉淀。**不持有 implementation surface**。
+
+---
+
+## 1. 角色定义 (dispatcher ≠ orchestrator)
+
+**peaks-solo 是一条 dispatcher skill**,跟 `peaks-code` 这种 orchestrator 平行存在,但更薄一层:
+
+| 角色 | 工作 | 是否持有 implementation surface |
+|---|---|---|
+| **dispatcher** (peaks-solo,本 skill) | 看用户 NL → 查 skill 池 → 转交 / 自规划 → 问沉淀 | **不持有**(只调度,不实现) |
+| **orchestrator** (peaks-code / peaks-content / peaks-doctor) | 看用户 NL → 走 11 步 runbook → 直接调 prd/rd/qa/ui/sc/txt 蜂群 → 沉淀 | **持有**(通过子蜂群间接持有) |
+
+**核心区分(HC-11 锁死):** peaks-solo 能看、能查、能转交、能跑通用工具(deep-search / WebSearch / Bash / Edit markdown)、能提议沉淀;**不能**写代码、**不能**写 PRD、**不能**跑 vitest、**不能**改 Loop Engineering 资产(那是用户主权)。
+
+**与 peaks-* leaf 的关系:** peaks-solo 是新加的第三条入口,不替代任何 leaf。3.x / 4.x 老用户继续 `/peaks-code` / `/peaks-content` / `/peaks-doctor` 等,**老入口 0 改动**(HC-10)。
+
+---
+
+## 2. 触发条件
+
+peaks-solo 由 LLM 在三种匹配类之一触发,**不依赖**任何固定的 slash 命令前缀:
+
+### 2.1 Source-trace 触发
+
+LLM 自身启动循环(如 Claude Code 启动新会话、resume 检测、auto-decide 路径)时,**没有**其它 skill 已被 dispatch,且用户描述诉求 → 走 peaks-solo 分诊。
+
+### 2.2 Trigger-phrase 触发(主路径)
+
+用户自然语言描述诉求 + 满足下列任一短语:
+
+- 自然语言描述诉求且无明确 peaks-* skill 选择(主触发)
+- "帮我处理这个" / "帮我看看"
+- "我不知道该用哪个"
+- "随便都行" / "你帮我决定"
+- "我有个事需要处理" / "我想做点东西"
+
+**反向:** 用户已明确指定 `/peaks-code` / `/peaks-content` / `/peaks-doctor` 等具体 skill → 不进 peaks-solo,直接 dispatch 到该 leaf(HC-10 老入口保留)。
+
+### 2.3 LLM-judge 触发
+
+用户 NL 表述是开放性、跨域、或者 LLM 无法判定属于哪个 peaks-* 域时(例如"帮我看下 GitHub trending top 10"),LLM 自己判断"这事没有专属 leaf",走 peaks-solo 兜底。
+
+---
+
+## 3. Triage 决策流
+
+peaks-solo 的 triage 流严格按以下伪代码执行:
+
+```
+1. user NL → peaks-solo 触发
+2. peaks-solo 跑 peaks skill search --query "<NL>" → 拿到 candidates []
+3. if candidates.length == 0:
+     → 走自规划兜底(§4)
+4. if candidates.length == 1:
+     → 直接 Skill tool 调起该 skill(转交,无感)
+5. if candidates.length >= 2:
+     → 走 AskUserQuestion 多选("NL 描述匹配到 N 个 peaks-* skill,请选一个:")
+     → 用户选 → Skill tool 调起
+     → 用户选 "(e) 都不对" → 自规划兜底(§4)
+6. leaf 跑完 → leaf 内部自己处理 sediment(不重复问;HC-9)
+7. 自规划跑完 → 走 §5 沉淀提议
+```
+
+### 3.1 多候选 AskUserQuestion 模板(HC-9 锁死)
+
+```
+question: "这个诉求匹配到以下 peaks-* skill,请选一个:"
+options:
+  - (a) <name + 一句话功能 + 适用场景>
+  - (b) <name + 一句话功能 + 适用场景>
+  - ...
+  - (e) 都不对,我自己跑
+multiSelect: false
+```
+
+**关键约束:** 必须是 AskUserQuestion,**不**允许让用户自由文本输入 skill 名(HC-9 Human-NL-Choice-Only)。
+
+### 3.2 用户主动说"我自己来" / "不用分诊"
+
+LLM **bail out**(返回控制权给用户,不调任何 leaf,不进自规划兜底)。
+
+### 3.3 关键词 → leaf skill 快速映射
+
+完整 ≥ 10 行关键词表见 `references/triage-decision-table.md`。下面只列最常见的 5 类:
+
+| keyword | → leaf skill |
+|---|---|
+| code / 改 bug / 全流程 / 端到端 | `/peaks-code` |
+| 文档 / article / 内容 / 写稿 | `/peaks-content` |
+| 健康 / 体检 / audit 报告 | `/peaks-doctor` |
+| issue / 批量修 / sweep | `/peaks-issue-fix-orchestrator` |
+| SOP / 流程 / 工作流 | `/peaks-sop` |
+
+---
+
+## 4. 自规划兜底
+
+当 `peaks skill search` 返回 0 个候选,或用户选"(e) 都不对",LLM 走自规划:用通用工具(deep-search / WebSearch / Bash / Edit markdown)完成用户诉求。
+
+### 4.1 允许工具清单
+
+| 工具 | 适用场景 | 限制 |
+|---|---|---|
+| `deep-search` skill | 信息查询 / 研究类(若已装) | 不主动 install |
+| `WebSearch` / `WebFetch` | 实时性高的查询(超过训练数据) | 内置工具 |
+| `Bash` | 数值计算 / 系统命令 / git 操作 | **不写 src/** 业务代码(那是 peaks-code 域) |
+| `Edit` / `Write` | 改 markdown / 改 yaml / 改 `.peaks/memory/` 文件 | **不改 src/** 业务代码;**不改 `skills/peaks-{code,content,doctor,...}/`** 任何 leaf skill 文件 |
+| `peaks memory extract` | 跑完提议沉淀 | 走现成 CLI,不重写 |
+
+完整 allowed + blocked 清单见 `references/fallback-tool-inventory.md`。
+
+### 4.2 自规划成功后的下一步
+
+自规划跑完 **必须** 走 §5 沉淀提议(HC-9 Human-NL-Choice-Only);不允许静默丢弃结果。
+
+---
+
+## 5. 沉淀提议
+
+自规划兜底跑完(leaf 转交跑完则由 leaf 内部 Step 11 处理,不重复问)→ peaks-solo 用 AskUserQuestion 提议沉淀。
+
+### 5.1 AskUserQuestion 模板(HC-9 锁死)
+
+```
+question: "本次结果是否沉淀?"
+options:
+  - (a) 沉淀为普通 lesson / convention → peaks memory extract --apply
+  - (b) 沉淀为 Loop Engineering 资产 → peaks asset crystallize
+  - (c) 改 scope 再沉淀(LLM 继续追问)
+  - (d) 不沉淀(单次性,丢弃)
+multiSelect: false
+```
+
+**默认推荐 = (a)**(普通 lesson / convention),LLM 必须给一段 NL 解释为什么推荐 (a) 或 (b),**不**允许默认推荐 (d)。
+
+完整 4-option 模板 + 触发条件矩阵见 `references/sediment-prompt-template.md`。
+
+### 5.2 触发条件矩阵
+
+| peaks-solo 跑完类型 | 是否问 | trigger |
+|---|---|---|
+| 转交 leaf 跑完 | **不问** | (leaf 内部 Step 11 处理) |
+| 自规划兜底跑完 | **默认问** | `success_default_prompt` |
+| 用户主动说"沉淀这个" | **立即问** | `user_explicit` |
+| LLM 识别 ≥ 2 个 reuse signals | **附带 4-section brief 问** | `llm_suggested` |
+
+---
+
+## 6. Out of scope (HC-11 锁死)
+
+peaks-solo 是 dispatcher,**不**是 orchestrator,**不**是 implementer。下面四类操作**禁止**直接在 peaks-solo 内部执行:
+
+- **no code** — 不写 / 不改 / 不重构 `src/**` 业务代码,也不调 peaks-rd / peaks-qa / peaks-ui / peaks-sc 子蜂群(那是 peaks-code 的 implementation surface)。遇到 code-domain 需求必须转交 `/peaks-code` 或建议用户自己跑。
+- **no PRD** — 不写产品需求文档(那是 peaks-code → peaks-prd 的范围)。peaks-solo 自己不持有 PRD 的生成 surface。
+- **no vitest** — 不跑测试,也不写 `tests/**` 业务测试。遇到测试需求必须转交 `/peaks-test` 或 `/peaks-code`。
+- **no Loop Engineering Asset mutation** — 不调 `peaks asset crystallize` 改 Loop Engineering 资产,也不写 `.peaks/standards/` / `.peaks/memory/` 之外的 peaks-loop 内部资产(那是用户主权,peak-* 技能层没有改动权)。
+
+**唯一例外:** 沉淀提议模板里 **选项 (b)** 提到 `peaks asset crystallize` 是给用户看的,不是 peaks-solo 自动执行;真正执行由用户在 AskUserQuestion 选中后由 LLM 代跑该 CLI。
+
+### 6.1 边角 case
+
+| case | 处理 |
+|---|---|
+| 用户中途换需求 | 重新跑 §3 triage,不复用旧 candidates |
+| 兜底工具失败(如 WebSearch 不可用) | 转下一种 allowed 工具;3 次失败后 AskUserQuestion 问用户是否要继续 |
+| 沉淀被拒后(用户选 d) | 不重复问,直接关闭本 skill session |
+| 用户在 AskUserQuestion 中输入自由文本(理论上不该发生) | 把它当作 NL 描述,重新跑 §3 triage |
+
+---
+
+## 相关引用
+
+- `references/triage-decision-table.md` — 关键词 → leaf skill 完整映射表(≥ 10 行)
+- `references/fallback-tool-inventory.md` — 自规划允许 + 禁止工具清单
+- `references/sediment-prompt-template.md` — 4-option AskUserQuestion 沉淀模板
+- spec: `docs/superpowers/specs/2026-07-08-peaks-solo-dispatcher-design.md` §3.1, §3.3, §3.4, §3.5
+- plan: `docs/superpowers/plans/2026-07-08-peaks-solo-dispatcher/s1-peaks-solo-skill.md`
+- decision memo: `.peaks/memory/user-decision-2026-07-08-revive-peaks-solo-as-dispatcher.md`

package/skills/peaks-solo/references/fallback-tool-inventory.md

@@ -0,0 +1,46 @@
+# peaks-solo — Fallback Tool Inventory (自规划兜底工具清单)
+
+> 当 `peaks skill search` 返回 0 候选,或用户在 AskUserQuestion 选"(e) 都不对",peaks-solo 走自规划兜底。本表规定自规划下 LLM **允许**与**禁止**调用的工具。
+> **核心边界(HC-11 锁死):** peaks-solo 不写业务代码、不写 PRD、不跑 vitest、不改 Loop Engineering 资产。下面 blocked 列表是 HC-11 的可执行形式。
+
+## Allowed tools (允许)
+
+| 工具 | 适用场景 | 限制 |
+|---|---|---|
+| `deep-search` skill | 信息查询 / 研究类 / 多跳事实核查 | 若 LLM CLI 平台装了 deep-search 自动出现;peaks-solo **不主动 install**;若未装,降级到 WebSearch |
+| `WebSearch` | 实时性高的查询(超过训练数据)/ trending / top N | 内置工具,无需 install |
+| `WebFetch` | 已知 URL 的内容抓取(单页,无 JS 渲染) | 内置工具;不爬取含登录 / 反爬的页面 |
+| `Bash` | 数值计算 / 系统命令 / git 操作 / `ls` / `cat` / `rg` / `grep` | 仅**只读命令**;不写 src/** 业务代码;不 rm;不 git push --force;不 chmod 777;不 curl \| sh |
+| `Edit` / `Write` | 改 markdown / 改 yaml / 改 `.peaks/memory/<user>/*.md` / 改临时文件 | **只允许**改:`*.md` / `*.yaml` / `*.yml` / `.peaks/memory/`;**不允许**改:`src/**` 业务代码 / `skills/peaks-{code,content,doctor,sop,status,test,resume,ide,final-review,slice-decompose,issue-fix-orchestrator,perf-audit,security-audit,audit}/**` 任何 leaf skill |
+| `peaks memory extract` | 跑完提议沉淀(选项 a) | 走现成 CLI,不重写;不创建新 CLI |
+| `AskUserQuestion` | 多候选分诊 / 沉淀提议 / 兜底方案选择 | HC-9 锁死;**不**引入自由文本输入 |
+
+## Blocked tools / operations (禁止)
+
+下表是**显式禁止**——LLM 在自规划兜底中**不能**调这些工具或跑这些操作。违反即违反 HC-11。
+
+| blocked 操作 | 原因 | 替代方案 |
+|---|---|---|
+| `rm -rf <path>` | 不可逆破坏性操作 | 改用 `mv <path> <path>.bak` 或只读 `ls` |
+| `git push --force` / `git push -f` | 改写历史,影响其他协作者 | 用 `git push --force-with-lease`(若必须)或拒绝 push 让用户自己来 |
+| `Edit src/**` / `Write src/**` 业务代码 | HC-11: no code;peaks-code 域 | 转交 `/peaks-code` |
+| `Edit skills/peaks-{code,content,doctor,sop,status,test,resume,ide,final-review,slice-decompose,issue-fix-orchestrator,perf-audit,security-audit,audit}/**` 任何文件 | HC-8 + HC-11: leaf skill 文件是用户主权 | 让用户自己改 |
+| `peaks asset crystallize`(直接调用) | Loop Engineering Asset mutation = 用户主权 | 仅在沉淀 AskUserQuestion 中作为**选项 (b)** 提及,不主动执行 |
+| `Edit .peaks/standards/**` / `Write .peaks/standards/**` | standards 是项目层 invariants | 让用户自己改 |
+| `Edit .claude-plugin/marketplace.json` | marketplace 是 install surface | 让用户自己改 / 转交 S2 |
+| `vitest` / `pnpm test` / 任何测试 runner | HC-11: no vitest | 转交 `/peaks-test` 或 `/peaks-code` |
+| `curl <untrusted-url> \| sh` / `wget \| bash` | RCE 风险 | 拒绝执行;若必须,先下载到文件,人工 review 后再执行 |
+| 写入用户 home 目录(`~/.*`) | 用户主权,peaks-solo 不动用户全局配置 | 让用户自己来 |
+
+## 边界判断原则
+
+LLM 拿不准"该不该做"时,按下面 3 条问自己:
+
+1. **这是 peaks-code 域吗?** 是 → 转交 `/peaks-code`。
+2. **这是用户主权领域吗?**(`src/**` 业务代码 / Loop Engineering Asset / 用户 home) → 让用户自己来。
+3. **这是 read-only 还是 mutating?** read-only(`ls` / `cat` / `rg` / WebSearch / WebFetch)→ OK;mutating 且不在 allowed 列表 → 拒绝。
+
+## 失败回退
+
+- 自规划下某个 allowed 工具失败 3 次(网络 / 权限 / not found)→ 走 AskUserQuestion 问用户:"这个工具一直失败,要不要换另一种?"(HC-9 锁死,不静默放弃)。
+- 自规划全部失败 → 走 AskUserQuestion:"peaks-solo 找不到合适工具,要不要: (a) 换更具体的 NL 重跑 / (b) 转人工 / (c) 结束"。

package/skills/peaks-solo/references/sediment-prompt-template.md

@@ -0,0 +1,62 @@
+# peaks-solo — Sediment Prompt Template (沉淀提议 AskUserQuestion)
+
+> 自规划兜底跑完(转交 leaf 跑完则由 leaf 内部 Step 11 处理,**不重复问**)→ peaks-solo 用本模板提议沉淀。
+> **核心约束(HC-9 Human-NL-Choice-Only 锁死):** 必须用 AskUserQuestion 4-option,**不**允许自由文本输入;**默认推荐 = (a)**,**不**是 (d)。
+
+## 4-option AskUserQuestion 模板
+
+```yaml
+question: "本次结果是否沉淀?"
+options:
+  - label: "(a) 沉淀为普通 lesson / convention"
+    description: "通过 peaks memory extract --apply 写到 .peaks/memory/<user>/,作为可复用 lessons/conventions。"
+  - label: "(b) 沉淀为 Loop Engineering 资产"
+    description: "通过 peaks asset crystallize 提升为 Loop Engineering Asset(标准 / skill / SOP / framework)。"
+  - label: "(c) 改 scope 再沉淀"
+    description: "本次 scope 不对,LLM 继续追问(例如想问清楚是什么 lesson),然后再走 (a) 或 (b)。"
+  - label: "(d) 不沉淀"
+    description: "单次性结果,丢弃,不写入任何 .peaks/ 路径。"
+multiSelect: false
+```
+
+## 默认推荐
+
+**默认推荐 = (a)**(普通 lesson / convention),**不**是 (d)。
+
+**为什么推荐 (a) 而不是 (b):** (b) 沉淀为 Loop Engineering Asset 是**重操作**——会进入 `.peaks/standards/` 或 `.peaks/memory/<asset>/`,影响所有 peaks-* skill 的语义。普通 lesson / convention 风险低、可回滚、不会污染项目 invariants,适合作为大多数一次性结果的首选归宿。(b) 仅在 LLM 识别 ≥ 2 个 reuse signals 时推荐。
+
+## LLM 必须给一段 NL rationale
+
+在调 AskUserQuestion 前,LLM **必须**先(在最终回复里,AskUserQuestion 之前)给一段 1-3 句的 NL 解释,说明为什么推荐 (a) 或 (b)。模板:
+
+> "我推荐 (a) 沉淀为普通 lesson / convention,理由:<1 句:本次结果的可复用程度 / 适用范围 / 为什么不是 (b) / 为什么不是 (d)>。如果你认为这是 Loop Engineering 级别的资产,选 (b);如果本次是单次性 / scope 不对,选 (c) 或 (d)。"
+
+**禁止:** 直接弹 AskUserQuestion 不解释;**禁止:** 默认推荐 (d)。
+
+## 触发条件矩阵
+
+| peaks-solo 跑完类型 | 是否问沉淀 | trigger 标记 |
+|---|---|---|
+| 转交 leaf 跑完 | **不问**(leaf 内部 Step 11 处理) | `leaf_handled` |
+| 自规划兜底跑完(成功) | **默认问** | `success_default_prompt` |
+| 自规划兜底跑完(部分失败) | **附带"是否继续"二选一 + 沉淀四选一**两个 AskUserQuestion | `partial_failure` |
+| 用户主动说"沉淀这个" / "把这个记下来" | **立即问**(跳过自规划完成检测) | `user_explicit` |
+| LLM 识别 ≥ 2 个 reuse signals(同一结果可在 ≥ 2 个未来场景复用) | **附带 4-section brief 问**(在 question 前先输出 brief:reuse contexts / scope / drift risks / crystallize recommendation) | `llm_suggested` |
+| 沉淀提议被用户选 (d) 后 | **不重复问**,直接关闭 peaks-solo session | `user_declined` |
+| 沉淀提议被用户选 (c) 后(scope 改) | **继续追问 → 再问一次 4-option** | `scope_change` |
+
+## 沉淀后的执行
+
+| 用户选 | LLM 执行 |
+|---|---|
+| (a) | `peaks memory extract --apply --session-id <sid> --project <repo>` —— 走现成 CLI,不重写 |
+| (b) | `peaks asset crystallize --asset <name> --project <repo>` —— 走现成 CLI;**仅**在用户明确选 (b) 后执行 |
+| (c) | 继续 AskUserQuestion 追问(例如"具体是什么场景的 lesson?" / "适用哪个子项目?") |
+| (d) | 不写任何文件;peaks-solo 直接结束 |
+
+## 反 pattern 警告
+
+- **不推荐 (b) 默认推荐** —— 大多数一次性结果进不了 Loop Engineering Asset 层;盲目 crystallize 会污染 `.peaks/standards/`。
+- **不静默丢弃** —— 用户没明示"不要沉淀",就必须问;选 (d) 必须是用户**主动选**,不是 LLM 默认。
+- **不重复问** —— 转交 leaf 跑完由 leaf 处理 sediment;peaks-solo 只对**自规划兜底跑完**问。
+- **不让用户自由文本** —— HC-9 锁死;不在 AskUserQuestion 选项之外加 free-form input。

package/skills/peaks-solo/references/triage-decision-table.md

@@ -0,0 +1,31 @@
+# peaks-solo — Triage Decision Table
+
+> 关键词 → leaf skill 快速映射。peaks-solo 在 §3 triage 决策流里先扫本表做"启发式预判",再用 `peaks skill search --query "<NL>"` 做精确匹配。
+> 匹配规则:case-insensitive substring(命中即行;同 NL 多行命中 → AskUserQuestion 多选)。
+
+| keyword | → | leaf skill |
+|---|---|---|
+| code / 代码 / 改 bug / 写功能 / 全流程 / 端到端 / PR / 测试覆盖 / 改 src | → | /peaks-code |
+| 文档 / article / blog / 内容 / 写稿 / 内容工作流 / 公众号 / 长文 | → | /peaks-content |
+| 健康 / health / 体检 / audit / 报告 / doctor report / 项目诊断 | → | /peaks-doctor |
+| issue / 修 issue / sweep / 批量修 / open issue 处理 | → | /peaks-issue-fix-orchestrator |
+| SOP / 流程 / 工作流 / 审批 / 门禁 / 标准操作 | → | /peaks-sop |
+| 性能 / perf / 慢 / 优化 / latency / 响应时间 | → | /peaks-perf-audit (or 自规划) |
+| 安全 / security / 漏洞 / vulnerability / OWASP | → | /peaks-security-audit (or 自规划) |
+| 状态 / status / 进度 / 现在到哪 / 当前在哪一步 | → | /peaks-status |
+| 跑测试 / test / vitest / unit test | → | /peaks-test |
+| 恢复 / resume / 继续 / 接上次 | → | /peaks-resume |
+| 不知道 / 随便 / 帮我 / 你帮我决定 / 我想做点东西 | → | peaks-solo triage (this skill) |
+| 切片 / slice / 拆解任务 / 大任务拆小 | → | /peaks-slice-decompose |
+| 想法 / brain-storm / 创意 / 探索思路 | → | /peaks-ide |
+| 最终审查 / final review / 验收 / 4-dim evidence | → | /peaks-final-review |
+| 研究 / research / 信息查询 / GitHub trending / top 10 / 调查 | → | (no leaf) → 自规划(deep-search / WebSearch) |
+| 一句话需求 / 一次性 / 单次性 | → | peaks-solo 自规划兜底 |
+
+## 使用说明
+
+1. **先扫表,再跑 search。** 表是 O(N) 的 substring 命中;跑 `peaks skill search` 是兜底。两者结果不一致时,以 search 为准(search 读的是 live frontmatter)。
+2. **多行命中 → AskUserQuestion 多选,**模板见 SKILL.md §3.1。
+3. **(e) "都不对" → 自规划兜底。** 走 SKILL.md §4。
+4. **空表条目:** "研究 / 信息查询 / GitHub trending / top 10" 等一次性信息查询,**当前 skill 池没有专属 leaf**,peaks-solo 直接走自规划(deep-search / WebSearch)。
+5. **本表是 opportunistic 表,**不是 exhaustive 表。S2/S3 后续可在 dogfood 中发现新关键词补行。