@damiangilgonzalez/harness 1.3.16 → 1.3.18

package/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "1.3.16",
+  "version": "1.3.18",
   "counts": {
     "skills": 294,
     "agents": 34,
@@ -3805,7 +3805,7 @@
     },
     {
       "id": "pr",
-      "description": "Use when writing a PR body: a visual summary, before/after evidence, and the merge danger (one-way or two-way door, blast radius).",
+      "description": "Use when writing a PR body or opening a PR: a visual summary, before/after evidence, the merge danger (one-way or two-way door, blast radius), and a title checked against the repo's CI title rule before `gh pr create`.",
       "tags": [
         "pull-request",
         "review",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@damiangilgonzalez/harness",
-  "version": "1.3.16",
+  "version": "1.3.18",
   "description": "A guided harness builder that turns your goal into a minimal, verified AI-agent setup using strict best practices.",
   "type": "module",
   "bin": {

package/scripts/doctor.js

@@ -261,7 +261,7 @@ export function doctor({ target, home, cwd }) {
   // #277 — the three "missing" lists share one shape, { id, path, action }, so whoever builds on top
   // of doctor reads them the same way; and `healthy` is what the exit code says.
   for (const [id, e] of Object.entries(state.skills)) {
-    for (const f of e.files) if (!existsSync(f)) report.missing.push({ id, path: f, action: 'Run `npx @ericrisco/rsc sync` to restore this managed skill.' });
+    for (const f of e.files) if (!existsSync(f)) report.missing.push({ id, path: f, action: 'Run `npx @damiangilgonzalez/harness sync` to restore this managed skill.' });
   }
   report.healthy = report.hookWired && !report.missing.length && !missingAgents.length && !missingCommands.length;
   return report;

package/scripts/rsc.js

@@ -885,11 +885,72 @@ async function main() {
       const root = W.resolveMainRoot(process.cwd());
       const sub = argv[1];
 
+      if (sub === 'new') {
+        // Creating is not cleaning, so the cleanup opt-out does not apply here. Two things a worktree
+        // was measured lacking when it is born: a current base (local main was 400 commits behind
+        // origin) and what git does not track (.env, dependencies). The base is fixed here; the rest
+        // is the project's own hook, because only the project knows what makes it runnable.
+        const branchAt = argv.indexOf('--branch');
+        const branch = branchAt >= 0 ? argv[branchAt + 1] : undefined;
+        if (branchAt >= 0 && (!branch || branch.startsWith('--'))) { say('rsc: --branch needs a branch name.'); process.exitCode = 1; return; }
+        const name = argv.slice(2).find((a, i, all) => !a.startsWith('--') && all[i - 1] !== '--branch');
+        if (!name) { say('Use: npx @damiangilgonzalez/harness worktrees new <nombre> [--branch <rama>]'); process.exitCode = 1; return; }
+        const out = W.createWorktree(root, name, { branch });
+        if (!out.created) { say(`rsc: ${out.reason}`); process.exitCode = 1; return; }
+        if (!out.fetched) say(`⚠️  git fetch origin failed (${out.fetchError || 'no detail'}); starting from the last known ${out.base}, which may be stale.`);
+        say(`✅ worktree ${out.path}`);
+        say(`   branch ${out.branch}, from ${out.base}`);
+        const hook = W.findPrepareHook(out.path, root);
+        if (hook) {
+          say(`   preparing it: ${hook}`);
+          const ran = W.runPrepareHook(hook, out.path, root);
+          if (!ran.ok) { say(`⚠️  the preparation hook failed (${ran.reason}). The worktree exists; finish preparing it by hand.`); process.exitCode = 1; }
+        } else {
+          say(`   No preparation hook (${W.PREPARE_HOOKS.join(' or ')}): .env files and dependencies are NOT copied — create the hook to do it on every new worktree.`);
+        }
+        say(`Next: cd "${out.path}"`);
+        return;
+      }
+
       if (!W.isCleanupEnabled(root)) {
         say(`worktree cleanup is off for this project (.rsc/${W.OPT_OUT}). Delete that file to turn it back on.`);
         return;
       }
 
+      if (sub === 'reap' && argv.includes('--merged')) {
+        // The bulk answer to "delete the worktrees" — asked four times in one project, with up to
+        // fourteen open. Merged is judged against origin's trunk AFTER a fetch, so a branch landed in
+        // the forge an hour ago counts. Without --confirm this only lists: one look, then one yes.
+        const fetched = W.fetchOrigin(root);
+        if (!fetched.ok) say(`⚠️  git fetch origin failed (${fetched.err || 'no detail'}); judging against the last known trunk.`);
+        const all = W.classifyWorktrees(root);
+        const merged = W.mergedCandidates(all);
+        const held = all.filter((c) => c.verdict === 'ask' && !merged.includes(c));
+        if (!merged.length) say('No merged worktree that can be removed in bulk.');
+        if (!argv.includes('--confirm')) {
+          if (merged.length) {
+            say(`${merged.length} merged worktree(s), clean, removable in one go:`);
+            for (const c of merged) say(`  ${c.path}${c.branch ? ` (${c.branch})` : ''}`);
+          }
+          for (const c of held) say(W.describe(c));
+          if (merged.length) say('\nRemove them all:  npx @damiangilgonzalez/harness worktrees reap --merged --confirm');
+          return;
+        }
+        for (const c of merged) {
+          const out = W.reapWorktree(root, c.path, { confirmed: true });
+          if (out.removed) {
+            say(`✅ removed ${out.path}`);
+            if (out.branchKept) say(`   branch ${out.branch} kept — git will not delete it safely, and while it exists the commits are recoverable.`);
+          } else {
+            say(`⏸  kept ${c.path} — ${out.reason}`);
+            process.exitCode = 1;
+          }
+        }
+        // What a bulk yes never covers, said out loud so nobody thinks it was forgotten.
+        for (const c of held) say(`⏸  kept ${c.path} — ${W.refusal(c)}`);
+        return;
+      }
+
       const candidates = W.classifyWorktrees(root).filter((c) => c.verdict !== 'skip');
 
       if (sub === 'reap') {
@@ -920,6 +981,7 @@ async function main() {
       say('');
       say('Remove the safe ones:  npx @damiangilgonzalez/harness worktrees reap');
       say('Remove one by name:    npx @damiangilgonzalez/harness worktrees reap <path> [--confirm]');
+      say('Remove every merged:   npx @damiangilgonzalez/harness worktrees reap --merged [--confirm]');
       return;
     }
     case 'capabilities': {
@@ -1195,7 +1257,7 @@ async function main() {
       return void (await runPurge(argv.includes('--dry-run'), argv.includes('--with-docs')));
     default:
       say(`rsc: unknown command '${cmd}'.`);
-      say('Use: npx @damiangilgonzalez/harness onboard | reassess | add <id...> | install --profile <p> | consult "<text>" | list | capabilities [--full|gap-log] | audit | registry refresh | doctor | sync | memory <on|off|status|save|resume|learn|metrics> | sello <on|off|status|…> | worktrees [reap [path] [--confirm]] | backups | restore <id|latest> | upgrade | repair | uninstall <id> | purge');
+      say('Use: npx @damiangilgonzalez/harness onboard | reassess | add <id...> | install --profile <p> | consult "<text>" | list | capabilities [--full|gap-log] | audit | registry refresh | doctor | sync | memory <on|off|status|save|resume|learn|metrics> | sello <on|off|status|…> | worktrees [new <nombre> [--branch <rama>] | reap [path|--merged] [--confirm]] | backups | restore <id|latest> | upgrade | repair | uninstall <id> | purge');
       say('Any command takes --target <claude|codex|cursor|copilot|gemini|…> (comma-separate for several)');
       say('   → without it, rsc uses the assistant already installed here; if two are, it asks instead of guessing.');
       process.exitCode = 1;

package/skills/building-agents/SKILL.md

@@ -14,7 +14,7 @@ A thin provider adapter, a disciplined agent loop, schema-validated tools, provi
 
 > Program against a **capability interface**, never a vendor SDK. Vendor specifics (model id, tool-schema shape, JSON mode, caching, token limits) live behind one adapter resolved from config. Model names and prices rot — if one appears in business logic it's a bug, and re-verify the dated tables before quoting a number.
 
-**Hand off instead when:** a new non-trivial feature has no approved spec + plan under `docs/wiki/sdd/` → stop and write one first (the `to-spec` skill, or `superpowers:writing-plans`), which routes back here once the plan is approved; one-line/low-risk changes go straight through. Anthropic-SDK internals (caching, thinking, batch) in a file that *only* imports `anthropic` → `claude-api` if your environment has it, since this skill stays multi-provider. Workspace scaffolding → [`harness`](../harness/SKILL.md). Choosing *which* coding agent to use → agent-eval territory. Pure prompt-wording tuning with no architecture change → prompt engineering, not this. A one-shot throwaway prompt, or no retrieval/tools/loop/evals at all → you don't need an agent; call the SDK directly and say so.
+**Hand off instead when:** a new non-trivial feature has no approved spec + plan under `docs/wiki/sdd/` → stop and write one first (the `to-spec` skill, then `to-tickets`), which routes back here once the plan is approved; one-line/low-risk changes go straight through. Anthropic-SDK internals (caching, thinking, batch) in a file that *only* imports `anthropic` → `claude-api` if your environment has it, since this skill stays multi-provider. Workspace scaffolding → [`harness`](../harness/SKILL.md). Choosing *which* coding agent to use → agent-eval territory. Pure prompt-wording tuning with no architecture change → prompt engineering, not this. A one-shot throwaway prompt, or no retrieval/tools/loop/evals at all → you don't need an agent; call the SDK directly and say so.
 
 ## Decision rules (read before writing code)
 

package/skills/code-review/SKILL.md

@@ -38,6 +38,8 @@ Busca el spec de origen, en este orden:
 
 Cualquier cosa en el repo que documente cómo debe escribirse el código, como `CODING_STANDARDS.md` o `CONTRIBUTING.md`. (En este proyecto: `CLAUDE.md` y `docs/CARTA-MAGNA.md`.)
 
+Si no encuentras ninguna, **no lo calles**: el eje de Estándares solo tendrá el smell baseline, y el reporte final debe decirlo en su primera línea ("Estándares: el repo no documenta estándares; solo se revisó el smell baseline"). Ofrece generar un `CODING_STANDARDS.md` inicial a partir de las convenciones que se observan en el repo (nombres, estructura, manejo de errores, tests). No lo escribas sin un sí del usuario.
+
 Además de lo que el repo documente, el eje de Estándares siempre lleva el **smell baseline** de abajo — un conjunto fijo de code smells de Fowler (_Refactoring_, cap. 3) que aplica incluso cuando el repo no documenta nada. Dos reglas lo condicionan:
 
 - **El repo manda.** Un estándar documentado del repo siempre gana; donde el repo avale algo que el baseline marcaría, suprime el smell.
@@ -74,7 +76,7 @@ Envía un único mensaje con dos llamadas a la herramienta `Agent`. Usa el subag
 - La ruta o el contenido recuperado del spec.
 - El encargo: "Reporta: (a) requisitos que el spec pedía y que faltan o están parciales; (b) comportamiento en el diff que no se pidió (scope creep); (c) requisitos que parecen implementados pero cuya implementación parece incorrecta. Cita la línea del spec por cada hallazgo. Menos de 400 palabras."
 
-Si falta el spec, sáltate el sub-agente de Spec y anótalo en el reporte final.
+Si falta el spec, sáltate el sub-agente de Spec y anótalo en el reporte final. Si faltan las fuentes de estándares, lanza igual el de Estándares con el baseline, y anótalo también junto con la oferta del `CODING_STANDARDS.md`.
 
 ### 5. Verificar los graves antes de presentarlos
 

package/skills/flujo-interfaz/SKILL.md

@@ -80,6 +80,13 @@ de las skills de diseno. Las mecanicas del framework son de `angular`, `react` o
 | Un flujo pierde usuarios a mitad | `leyes-de-retencion` -> `mejor-redaccion` -> `mejor-accesibilidad` |
 | Solo pulir lo que ya hay | `revision-interfaz` |
 
+## Cuando cuenta como hecho
+
+Un cambio visual no esta hecho sin una **captura del entorno donde el usuario lo va a probar**
+(su navegador, su URL, su viewport; no un test ni un render aislado si el usuario no prueba ahi).
+Que compile y pasen los tests no dice nada de como se ve. Si no puedes abrirlo, dilo y no lo
+declares terminado: "no lo he mirado en un navegador" es un trabajo sin acabar, no un matiz.
+
 ## Donde encaja en la metodologia
 
 Esta cadena vive **dentro** del paso de construir, no lo sustituye. El trabajo sigue llegando con

package/skills/grill-with-docs/SKILL.md

@@ -2,7 +2,6 @@
 name: grill-with-docs
 description: "A relentless interrogation to sharpen a plan or design, which also builds documentation (ADRs and a glossary) along the way."
 tags: ["planning", "questioning", "documentation", "adr"]
-disable-model-invocation: true
 profiles: [core, ui, full]
 origin: damiangil
 ---

package/skills/harness/references/daily-curation-automation.md

@@ -42,7 +42,8 @@ process inbox/, then scan the workspace (minus .rscignore) for un-ingested docum
 applying Clean-as-you-go (move loose root-level + inbox files into raw/ after a
 verified ingest; copy files nested in user-maintained folders and propose
 consolidation), record them in wiki/.ingested.json, and list ambiguous folders as
-proposals (do not grab them). Then compile pending raw/worklog/ material into wiki/
+proposals (do not grab them). Then compile pending raw/worklog/ material (skip this
+step if raw/worklog/ does not exist; do not list a missing folder) into wiki/
 pages (update existing before creating new), keep frontmatter + relative markdown
 links OKF-consistent, append significant decisions to wiki/harness/decisions.md,
 refresh the .base-backed navigation, recompute scores.json + the score: property,

package/skills/harness/references/wiki-protocol.md

@@ -417,7 +417,9 @@ safety rule the curation pass honors.
 ### Steps
 
 1. **Capture.** Write the worklog to `raw/worklog/YYYY-MM-DD-<slug>.md`
-   (`status: unprocessed`). Infer `topic` from content.
+   (`status: unprocessed`). Infer `topic` from content. If `raw/worklog/` does
+   not exist yet, create it (`mkdir -p`) — never list it blindly: the checkpoint
+   hook already says whether it exists and how many worklogs it holds.
 2. **Compile.** Run the normal Compile (Fetch is a no-op — the worklog is already
    `raw`): distill durable points into `wiki/<topic>/` articles (update existing
    before creating new), add markdown-link cross-refs + `## Related`, cascade.

package/skills/implement/SKILL.md

@@ -2,7 +2,6 @@
 name: implement
 description: "Implements a piece of work from a spec or a set of tickets."
 tags: ["implementation", "spec", "tickets"]
-disable-model-invocation: true
 profiles: [core, ui, full]
 origin: damiangil
 ---
@@ -14,6 +13,8 @@ Usa TDD (superpowers:test-driven-development) donde sea posible, en las costuras
 
 Ejecuta el typechecking con regularidad, archivos de test sueltos con regularidad, y la suite de tests completa una vez al final.
 
+Si el trabajo cambia algo visible, no está hecho hasta que lo hayas mirado: adjunta una captura del entorno donde el usuario lo va a probar. Si no puedes abrirlo, dilo y no lo des por terminado.
+
 Al terminar, usa /code-review para revisar el trabajo.
 
 Commitea tu trabajo en la rama actual.

package/skills/implement-spec/SKILL.md

@@ -2,12 +2,11 @@
 name: implement-spec
 description: "Implement the result of /to-spec and /to-tickets in code, orchestrating parallel implementer subagents over the ticket graph."
 tags: ["implementation", "spec", "tickets", "subagents", "orchestration"]
-disable-model-invocation: true
 profiles: [core, ui, full]
 origin: damiangil
 ---
 
-Te han dado una spec. Esa spec debería tener tickets asociados que describen cómo implementarla.
+Te han dado una spec. Esa spec debería tener tickets asociados que describen cómo implementarla. Si no los tiene, **para**: lanza `to-tickets` sobre la spec y vuelve aquí con el grafo ya publicado. Sin tickets no hay frontera que repartir, y esto se convierte en una implementación a ciegas.
 
 El issue tracker debería estar descrito en `docs/agents/issue-tracker.md`. Si falta, pregunta al usuario dónde viven los tickets y cómo se cierra el trabajo (PR, issue, fichero).
 
@@ -30,9 +29,11 @@ Lanza los **subagentes implementadores** en segundo plano siempre que se pueda,
 4. Usa **subagentes implementadores** para implementar cada ticket, cada uno en su propio worktree y en su propia rama (`superpowers:using-git-worktrees`). Cada subagente implementador:
    - confirma que su worktree parte de la rama de integración antes de empezar, y se reinicia sobre ella si no;
    - llama a la herramienta Skill con `superpowers:test-driven-development` para construir el ticket;
-   - fusiona la punta de la rama de integración en su propia rama antes de decir que ha terminado.
+   - fusiona la punta de la rama de integración en su propia rama antes de decir que ha terminado;
+   - si el ticket cambia algo visible, adjunta una captura del entorno donde el usuario lo va a probar;
+   - entrega un informe final de **150 palabras como máximo**: qué cambió, ficheros, resultado de los tests y bloqueos. El detalle va al commit o al ticket, no a la respuesta. Pon este tope literal en el prompt de cada subagente: cada entrega entra entera en tu contexto, y decenas de entregas largas te hacen olvidar lo que acaba de decir el usuario.
 
-5. Cuando un **subagente implementador** termina, fusiona su trabajo en la rama de integración con un **subagente fusionador**.
+5. Cuando un **subagente implementador** termina, revisa su entrega antes de fusionar. Si toca algo visible y no trae captura, o dice que no lo ha mirado en un navegador, **no está hecho**: devuélvesela para que lo mire. Si la entrega es buena, fusiona su trabajo en la rama de integración con un **subagente fusionador**.
 
 6. Si eso cambia la **frontera** de tickets disponibles, lanza más **subagentes implementadores** para los tickets nuevos. Así se mantiene la máxima concurrencia.
 

package/skills/init/SKILL.md

@@ -165,11 +165,13 @@ means the user starts clean.
    machine and flags overlap or skills with no footprint, so the project starts with the right set
    rather than a pile. Summarize at their level.
 4. **Tell them about the danger guard.** A `technical_level` of `non-technical` or `mixed` (and the
-   state before any profile exists) turns on a `PreToolUse` guard that blocks irreversible commands —
-   `rm -rf`, `git push --force`, `git reset --hard`, `DROP`/`TRUNCATE`, `DELETE`/`UPDATE` with no
-   `WHERE`, `dd` to a device, `curl | bash` — and asks for a safer alternative. A fully `technical`
-   user is never guarded. It turns off only if the user explicitly asks: `.rsc/.no-danger-guard`.
-   Mention it when you set a non-technical level, so a later block is not a surprise.
+   state before any profile exists) turns on a `PreToolUse` guard for irreversible commands —
+   `rm -rf`, `git push --force`, `git reset --hard`, `git branch -D`, `DROP`/`TRUNCATE`,
+   `DELETE`/`UPDATE` with no `WHERE`, `dd` to a device, `curl | bash`. For `non-technical` (or no
+   profile) Claude Code asks the user to confirm each one, saying what the command destroys; their
+   yes is the order that runs it. For `mixed` it only warns. A fully `technical` user is never
+   guarded. It turns off only if the user explicitly asks: `.rsc/.no-danger-guard`.
+   Mention it when you set a non-technical level, so a later confirmation is not a surprise.
 
 ### Phase 5 — HANDOFF
 

package/skills/orient/SKILL.md

@@ -63,6 +63,9 @@ The dial is spoken, not configured: when the user says "explícame más / menos"
 tanto" or "enséñame", update `accompaniment_level` in the profile, confirm in one line, and apply
 the new depth from this turn on.
 
+When a front closes (PR merged or opened, ticket done), `➡️` offers `handoff` and a fresh session
+before more work in this one; see the contract.
+
 At a real fork, ask. Deciding alone is faster and wrong — the fork is the moment the person's
 judgment is worth more than yours.
 

package/skills/orient/references/orientation-contract.md

@@ -28,6 +28,15 @@ Read `docs/wiki/harness/user-profile.md`. Two fields combine: `accompaniment_lev
 
 `technical_level` is orthogonal (plain language for `non-technical` even at L0; skip 101s for `technical` even at L3). Missing profile → assume `non-technical` + **L3** and offer to set the dial.
 
+## When a front closes
+
+A front ("frente") closes when its PR is merged or opened, or its ticket is done. That turn's `➡️`
+proposes `handoff` and a new session as the first option, instead of starting the next front here.
+A session that runs for days through several `/compact` loses what the user just said: each
+compaction and each subagent hand-back pushes the latest instructions further out. A handoff
+document carries what matters; the new session starts with room for it. If the user prefers to
+continue, continue — offer once per front, not every turn.
+
 ## Division of labor
 
 - `orient` guides the **person** (this contract).

package/skills/postgresdb/SKILL.md

@@ -13,7 +13,7 @@ fix slow SQL, run zero-downtime migrations, and operate/secure the database. Too
 example is runnable.
 
 New, non-trivial feature with no approved spec + plan under `docs/wiki/sdd/`? Hand off to
-the `to-spec` skill before writing feature code (or `superpowers:writing-plans`); build
+the `to-spec` skill before writing feature code (then `to-tickets`); build
 straight from here only for a genuinely one-line, low-risk change.
 
 Deep dives: [schema-and-indexing](references/schema-and-indexing.md) (types, constraints, every index

package/skills/pr/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: pr
-description: "Use when writing a PR body: a visual summary, before/after evidence, and the merge danger (one-way or two-way door, blast radius)."
+description: "Use when writing a PR body or opening a PR: a visual summary, before/after evidence, the merge danger (one-way or two-way door, blast radius), and a title checked against the repo's CI title rule before `gh pr create`."
 tags: ["pull-request", "review", "evidence", "risk"]
 profiles: [core, ui, full]
 origin: damiangil
@@ -171,3 +171,17 @@ La evidencia basada en ejecución va justo detrás: resultados de tests, salida
 Di si es una puerta de un sentido o de dos sentidos. Por una puerta de dos sentidos se puede volver atrás; por una de un sentido, no. Una PR que es barata de revertir tiene menos riesgo. Los cambios que implican acciones destructivas o decisiones difíciles de revertir son puertas de un sentido.
 
 El radio de impacto es el alcance posible de los efectos que introduce la PR. Considera todas las posibilidades. Ejemplos: saltos de layout, roturas para quien consume el código, el comportamiento en móvil, etc.
+
+## Título
+
+Antes de `gh pr create`, comprueba que el título pasa el check de CI del repo. Un título mal formado tumba la PR aunque el código esté bien.
+
+1. Busca la regla: `grep -rnE "semantic-pull-request|pull_request\.title|commitlint" .github/workflows/` y, si aparece commitlint, su configuración (`commitlint.config.*`, `.commitlintrc*`, `package.json`). En `amannn/action-semantic-pull-request` mira `types`, `scopes`, `requireScope` y `subjectPattern`.
+2. Saca la regex o los tipos permitidos y prueba el título:
+
+   ```bash
+   node -e 'const re=/^(feat|fix|docs|refactor|test|chore)(\([\w-]+\))?!?: .+/u; const t=process.argv[1]; console.log(re.test(t)?"OK":"FALLA: "+t)' "fix(pr): valida el título"
+   ```
+
+   Si el repo antepone un gitmoji u otro prefijo, inclúyelo en la regex tal como lo pida la regla.
+3. Si falla, corrige el título antes de crear la PR. Si no hay ningún check de título, sigue la convención de los commits recientes (`git log --oneline -10`).

package/skills/pr/evals/cases.yaml

@@ -11,6 +11,8 @@ should_trigger:
     why: "Disparador directo de pr: peligro de fusión y radio de impacto."
   - prompt: "¿Qué pongo en la PR para que se revise rápido?"
     why: "Disparador directo de pr."
+  - prompt: "Abre la PR, pero que el título no vuelva a tumbar el CI."
+    why: "Disparador directo de pr: el título se valida contra el check del repo."
 
 should_not_trigger:
   - prompt: "Revisa el código de esta PR."
@@ -31,3 +33,7 @@ capability:
     must_include:
       - "Peligro de fusión"
       - "Radio de impacto"
+  - scenario: "El repo tiene un check de título de PR en sus workflows."
+    must_include:
+      - ".github/workflows"
+      - "gh pr create"

package/skills/suggest/SKILL.md

@@ -34,8 +34,8 @@ The shape of the chain:
   when the effort will not fit one sitting.
 - Once the shape is known, capture it as a spec, and give any call that would be costly to undo
   its own ADR.
-- Build from a written plan. Tickets are the exception: only work that grew out of the
-  wayfinding step is broken into them first. A screen also routes through `flujo-interfaz`.
+- Small work builds straight from the clarified ask; big work is split into tickets first and
+  built from them. A screen also routes through `flujo-interfaz`.
 - Review the finished change against the repo's standards and against the spec before merging.
 
 Two exceptions, and say out loud when you take one:
@@ -54,9 +54,9 @@ whole run — advance through the phases without re-asking.
 If a step's skill is not installed, offer to add it (§2) before routing. `implement` always
 pairs with test-driven development and a verification pass before anything is called done.
 
-Exact skill names and order: `grill-with-docs` or `wayfinder` -> `superpowers:brainstorming` or
-`to-spec` (and `write-adr` for decisions) -> `superpowers:writing-plans` ->
-`superpowers:executing-plans` (big spec: `to-tickets` -> `implement-spec`) -> `code-review` -> `pr`.
+Exact skill names, by size. Small: `superpowers:brainstorming` or `grill-with-docs` ->
+`implement`. Big: `grill-with-docs` or `wayfinder` -> `to-spec` -> `to-tickets` ->
+`implement-spec`. Both: `write-adr` for decisions, then `code-review` -> `pr`.
 On Claude Code this rule also arrives as a per-turn hook; the brevity here is deduplication,
 not relaxation.
 

package/skills/to-spec/SKILL.md

@@ -2,7 +2,6 @@
 name: to-spec
 description: "Turns the current conversation into a spec (PRD) and publishes it to the project's issue tracker — no interview, just a synthesis of what has already been discussed."
 tags: ["spec", "prd", "issue-tracker"]
-disable-model-invocation: true
 profiles: [core, ui, full]
 origin: damiangil
 ---

package/skills/to-tickets/SKILL.md

@@ -2,7 +2,6 @@
 name: to-tickets
 description: "Breaks down a plan, a spec, or the current conversation into a set of \\"tracer bullet\\" tickets (vertical slices), each declaring which tickets block it, and publishes them to the configured tracker — as local files (one per ticket) or as issues with native blocking links in a real tracker."
 tags: ["tickets", "planning", "issue-tracker"]
-disable-model-invocation: true
 profiles: [core, ui, full]
 origin: damiangil
 ---

package/skills/wayfinder/SKILL.md

@@ -2,7 +2,6 @@
 name: wayfinder
 description: "Plans a huge chunk of work — more than fits in a single agent session — as a shared map of decision tickets in the repo's issue tracker, and resolves them one at a time until the path to the destination is clear."
 tags: ["planning", "issue-tracker", "large-scale-work"]
-disable-model-invocation: true
 profiles: [core, ui, full]
 origin: damiangil
 ---

package/targets/_md-block.js

@@ -21,7 +21,7 @@ export function writeSkill(id, fromDir, toPath) {
 export function wireHook(paths, sourceMd, policy = {}) {
   const full = stripFrontmatter(readFileSync(sourceMd, 'utf8'));
   const body = policy.codeHooks === false
-    ? `# rsc-suggest — always-on operations layer\n\nRead \`docs/wiki/harness/user-profile.md\` before acting. If the user wants to build, add or change something, route it through \`grill-with-docs\` or \`wayfinder\`, then write the spec with \`superpowers:brainstorming\` or \`to-spec\` (decisions go to \`write-adr\`) before any code, then build it with \`superpowers:writing-plans\` (or \`to-tickets\` then \`implement-spec\` for a big spec or one that came out of \`wayfinder\`), and review it with \`code-review\` before merging, writing the PR body with \`pr\` (si toca pantalla, \`flujo-interfaz\` dice que skill de diseno manda) — unless it is a one-line change, typo, or config bump, which skips the chain. Use \`orient\` to keep the user situated and \`suggest\` to offer a missing skill only when the current task needs it. Close with the configured orientation block.\n`
+    ? `# rsc-suggest — always-on operations layer\n\nRead \`docs/wiki/harness/user-profile.md\` before acting. If the user wants to build, add or change something, pick the chain by size and chain each step yourself. Small (one sitting, one ticket): clarify with \`superpowers:brainstorming\` (idea still vague) or \`grill-with-docs\` (a plan to harden), then build with \`implement\`. Big: \`grill-with-docs\` or \`wayfinder\` -> \`to-spec\` -> \`to-tickets\` -> \`implement-spec\`. Hard-to-reverse decisions go to \`write-adr\`; review with \`code-review\` and write the PR with \`pr\` before merging (si toca pantalla, \`flujo-interfaz\` dice que skill de diseno manda) — unless it is a one-line change, typo, or config bump, which skips the chain. Use \`orient\` to keep the user situated and \`suggest\` to offer a missing skill only when the current task needs it. Close with the configured orientation block.\n`
     : full;
   const block = `${MARK_START}\n${body}\n${MARK_END}`;
   let doc = existsSync(paths.hookTarget) ? readFileSync(paths.hookTarget, 'utf8') : '';

package/targets/claude.js

@@ -9,6 +9,15 @@ const HERE = dirname(fileURLToPath(import.meta.url));
 // so a `Bash`-only matcher let `git push --force` or a commit without gitmoji through unasked.
 const SHELL_TOOLS = 'Bash|PowerShell';
 
+// Mods de la terminal de Claude Code que el harness deja activados en cada proyecto.
+// `cache-tax` mantiene caliente la caché del prompt en las pausas, y eso gasta tokens.
+export const CLAUDE_CODE_MODS = Object.freeze([
+  { id: 'skins@hellosverre-mods', marketplace: 'hellosverre-mods', repo: 'hellosverre/claude-skins' },
+  { id: 'savvy-progress@claude-kit', marketplace: 'claude-kit', repo: 'johnnyvizz/claude-kit' },
+  { id: 'filetree@claude-code-filetree', marketplace: 'claude-code-filetree', repo: 'data-goblin/claude-code-filetree' },
+  { id: 'cache-tax@claude-code-mods', marketplace: 'claude-code-mods', repo: 'karanb192/claude-code-mods' },
+]);
+
 export function writeSkill(id, fromDir, toPath) {
   // Migrate away from the legacy nested layout (.claude/skills/rsc/<id>) that
   // Claude Code never discovered — it only reads .claude/skills/<name>/SKILL.md.
@@ -166,7 +175,8 @@ export function wireHook(paths, sourceMd, policy = {}) {
 
   if (policy.codeHooks !== false) {
   // Ship guard: a PreToolUse(Bash) hook that DENIES switching to / merging the trunk
-  // while the current feature branch has uncommitted or unpushed work — forcing the
+  // while the current feature branch has unpushed commits (uncommitted changes only get a
+  // worktree hint: they may be the user's own) — forcing the
   // commit → push → PR close (the `ship` skill). Materialized + node-run (Windows-safe),
   // registered idempotently, fail-open, opt-out via .rsc/.no-ship-guard. Other
   // (non-rsc) PreToolUse hooks are preserved.
@@ -235,6 +245,16 @@ export function wireHook(paths, sourceMd, policy = {}) {
     for (const id of policy.plugins) settings.enabledPlugins[id] = true;
   }
 
+  // Los mods de Claude Code van en cada install y update, se use el flujo o no. Con el marketplace
+  // en `extraKnownMarketplaces`, Claude Code ofrece instalarlos al abrir el proyecto. Un `false`
+  // que alguien puso a mano es una decisión suya y se respeta.
+  settings.extraKnownMarketplaces ||= {};
+  settings.enabledPlugins ||= {};
+  for (const { id, marketplace, repo } of CLAUDE_CODE_MODS) {
+    settings.extraKnownMarketplaces[marketplace] ||= { source: { source: 'github', repo } };
+    if (settings.enabledPlugins[id] !== false) settings.enabledPlugins[id] = true;
+  }
+
   // Danger guard: its own decision, not part of the code-hook block above (#273). `init` promises it
   // to non-technical and mixed users whatever the project is, and a plan that predates the field is
   // read as "yes": a non-technical operations install gets the guard it was missing on its next sync,