@gtrabanco/pi-agentic-workflow 0.10.4 → 0.10.6

package/README.md

@@ -1,7 +1,5 @@
 # @gtrabanco/pi-agentic-workflow
 
-> 🇪🇸 [Versión en español](README.es.md)
-
 One install of the [agentic-workflow](https://github.com/gtrabanco/agentic-workflow)
 method into [Pi](https://github.com/badlogic/pi-mono): the canonical skills, a
 friendly slash command for each of them, and optional per-command model routing

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gtrabanco/pi-agentic-workflow",
-  "version": "0.10.4",
+  "version": "0.10.6",
   "description": "Pi package: canonical agentic-workflow skills, friendly slash commands, and per-command model routing.",
   "keywords": [
     "pi-package",
@@ -32,7 +32,6 @@
     "skills",
     "package.json",
     "README.md",
-    "README.es.md",
     "LICENSE"
   ],
   "engines": {

package/skills/design-feature/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: design-feature
 user-invocable: true
-version: 3.3.0
+version: 3.4.0
 argument-hint: <idea | NN-slug> [<instruction>]
 author: "Gabriel Trabanco <gtrabanco@users.noreply.github.com>"
 license: MIT
@@ -89,15 +89,18 @@ another `references/` path; in the DEFER column, bare names live in the same
 directory as the links.
 
 **Hard stop for an incomplete raw-idea interview:** LOAD exactly
-`references/INTERVIEW.md` and no other reference. Ask its one next
-question, return `NEEDS_INPUT`, and end the turn. `WRITE_AND_UPSERT.md` is
+`references/INTERVIEW.md` and no other reference. Present its ONE compact
+form-turn — the ≤ 6 fixed rubric slots plus the identity rows, each carrying a
+recommended default the user can accept with one word — then return
+`NEEDS_INPUT` and end the turn when slots remain unresolved; genuine ambiguity
+gets at most **2 follow-up turns**, never a third ask. `WRITE_AND_UPSERT.md` is
 forbidden until every mandatory interview slot is resolved; that resource then
 owns closure rows and writing.
 
 | Condition now | LOAD now | DEFER / SKIP now |
 |---|---|---|
 | Bare existing slug, no instruction | [interview](references/INTERVIEW.md) through its interaction rule; report status and stop | `WRITE_AND_UPSERT.md`, `UPSERT_EXAMPLE.md`, `REPAIR.md`, `PORTABILITY.md` |
-| Brand-new idea with any mandatory interview slot unresolved | [interview](references/INTERVIEW.md) only; ask exactly its next question and stop | `WRITE_AND_UPSERT.md`, `UPSERT_EXAMPLE.md`, `REPAIR.md`, `PORTABILITY.md` |
+| Brand-new idea with any mandatory interview slot unresolved | [interview](references/INTERVIEW.md) only; present exactly one form-turn (≤ 6 slots + identity, each with a default) and stop | `WRITE_AND_UPSERT.md`, `UPSERT_EXAMPLE.md`, `REPAIR.md`, `PORTABILITY.md` |
 | New idea after every mandatory interview slot resolves | [interview](references/INTERVIEW.md), then [closure, write, and upsert](references/WRITE_AND_UPSERT.md) | `UPSERT_EXAMPLE.md` unless shape is ambiguous; `REPAIR.md`, `PORTABILITY.md` |
 | Existing slug plus instruction | interview, then closure/write/upsert | [upsert example](references/UPSERT_EXAMPLE.md) unless shape is ambiguous; `REPAIR.md`, `PORTABILITY.md` |
 | Existing slug whose `progress.md` carries `SPEC-REVIEW-FAIL` or `NEEDS-DESIGN` | [interview](references/INTERVIEW.md), then [review repair](references/REPAIR.md) | `UPSERT_EXAMPLE.md`, `PORTABILITY.md` |

package/skills/design-feature/references/INTERVIEW.md

@@ -16,10 +16,13 @@
 3. **Raw-idea interview (folded in, only when starting from zero or the
    instruction leaves genuine gaps).** Fixed protocol — structural, not
    judgement:
-   - **One question per turn, never batched.** Each question carries a
-     recommended default the user can accept with one word. Ask nothing the
-     docs or the instruction already answer.
-   - **Vagueness rubric (fixed slots — the question list IS this list).**
+   - **One form-turn, then at most 2 follow-up turns.** The first interview
+     turn is ONE compact form covering every rubric slot below plus the
+     identity slots, each row carrying a recommended default the user accepts
+     with one word (or edits in place). Ask nothing the docs or the instruction
+     already answer. Genuine ambiguity the form cannot resolve gets at most
+     **2 follow-up turns** — never a third ask.
+   - **Vagueness rubric (fixed slots — the form rows ARE this list).**
      Probe each slot until it is filled or explicitly `n/a: <reason>`:
      1. **Affected users/roles** — who uses it; who must not.
      2. **Error & edge states** — what happens on failure / empty / invalid.
@@ -40,11 +43,13 @@
    - **Escalation (structural).** If, after the interview, **≥ 3 rubric slots
      remain empty** (neither filled nor `n/a`), do not guess: end the turn
      `NEEDS_INPUT`, listing the empty slots verbatim as the pending
-     questions — the feature is not designable yet.
-   - The identity questions ride the same one-per-turn protocol: problem &
-     goal, business goals, size estimate (`XS/S/M/L` — XS/S stays SPEC-only,
-     M/L gets the full artifact set), non-goals / future work, traceability
-     (offer a tracking issue; if created, the eventual PR will `Closes #n`).
+     questions — the feature is not designable yet. Ambiguity that survives the
+     follow-up budget escalates the same way (`interview:ambiguity-cap`): the
+     2-turn cap never licenses a third ask.
+   - The identity questions ride the same form-turn: problem & goal, business
+     goals, size estimate (`XS/S/M/L` — XS/S stays SPEC-only, M/L gets the
+     full artifact set), non-goals / future work, traceability (offer a
+     tracking issue; if created, the eventual PR will `Closes #n`).
 4. **Proportional research.** Capability closure (step 5) is cheap and comes
    first. Reach for external or domain research **only** when the feature
    touches a domain genuinely new to the project (a regulation, an unfamiliar

package/skills/orchestration-envelope/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: orchestration-envelope
 user-invocable: false
-version: 2.0.2
+version: 2.1.0
 author: "Gabriel Trabanco <gtrabanco@users.noreply.github.com>"
 license: MIT
 description: >

package/skills/orchestration-envelope/references/TURN_CONTRACT.md

@@ -20,6 +20,37 @@
 
 **Push policy supplement:** Box 4 covers the end-of-unit push. Some consumers (e.g. `execute-phase/references/PREFLIGHT.md`) define additional push rules for mid-phase commits when a PR is already open — those per-consumer supplements extend this contract and must be loaded alongside it by the consumer skill.
 
+## Machine-check profile (boxes 1–5)
+
+Boxes 1–5 are mechanical. Demonstrate them by running the verifier — the
+scaffold shim `template/.agentic-workflow/hooks/turn-contract.sh` (bash + git +
+gh; presence-only box2 — the conditional phase-lint clause is engine-only, so
+a shim receipt leaves that clause unverified, ED-52-3) or the producer-crate
+engine `packages/agentic-workflow/bin/turn-contract.mjs` — and pasting its
+one-line receipt beside the git/gh evidence. When the receipt is pasted,
+reciting boxes 1–5 in prose is **not** required. Verifier unavailable (no bash,
+or no shim/engine in the project)? Recite boxes 1–5 exactly as before — the
+profile never weakens a box, and a turn with neither receipt nor recitation is
+unverified. Boxes 6–11 are unchanged: they stay agent-attested judgment calls
+no script may speak for. The closing `→ Next:` block may be echoed from the
+`workflow-status` envelope instead of hand-authored: `next.recommended`
+supplies the recommended line and `next.alternatives` the `·` sub-bullets,
+preserving the fixed block shape.
+
+```text
+turn-contract-receipt@1
+ok-line:    TURN-CONTRACT ok
+fail-line:  TURN-CONTRACT fail box<N>: <code>
+codes:      branch-default | not-a-repo | no-commits | acceptance-missing |
+            phase-lint-failed | pr-not-open | pr-head-mismatch |
+            pr-unreachable | dirty-tree | ahead-of-remote
+exit:       0 ok | 1 contract fail | 2 usage error
+order:      boxes 1→5, first failure wins; within box5, dirty-tree precedes
+            ahead-of-remote
+stdout:     exactly one line; diagnostics never (git/gh evidence stays in the
+            agent's transcript)
+```
+
 ## Hand-off grammar (versioned — read by `scripts/normative-drift.test.mjs`)
 
 Box 9's closing hand-off and box 10's machine result are ordered here, so the
@@ -51,5 +82,6 @@ next | recommended
 next | alternatives
 next | tier
 next | suggested
+next | continuation
 ```
 

package/skills/review-change/references/PERSIST_AND_DECIDE.md

@@ -1,5 +1,9 @@
 ## Persist and decide
 
+When the unit's review receipt is missing or stale, the re-entry command is the
+**emitted** `next.continuation` (`rendering` for display) — quote it, never author
+exact command tokens.
+
 11. **Persist fix-now findings to the fold ledger.** The unit's **fix-now fold
    ledger** is `review-findings.md`, located beside its other docs
    (`docs/features/<NN>-<slug>/review-findings.md` for a feature,

package/skills/review-plan/references/OUTPUT.md

@@ -6,7 +6,10 @@ Append the receipt to `docs/features/<NN>-<slug>/progress.md` (fix units:
 `docs/fix/<N>/progress.md`), then append every finding row to
 `planning-findings.md` (ledger contract: `pre-execution-review/references/LEDGERS.md`).
 Any later write to a reviewed plan artifact rotates `artifactRevisionId` and makes
-this receipt stale — that is the contract working, not a mistake.
+this receipt stale — that is the contract working, not a mistake. When a missing or
+stale receipt is the reason for re-entry, the re-entry command is the **emitted**
+`next.continuation` (`rendering` for display): quote it, never author exact command
+tokens.
 
 One `PreExecutionReviewReceipt v1`
 (`agentic-workflow/pre-execution-review-receipt@1`) per review:

package/skills/review-spec/references/OUTPUT.md

@@ -8,7 +8,9 @@ each finding row to the unit's `planning-findings.md`** with
 create the file from that contract when the unit has none. A findings ledger the
 reviewer cannot write is not a ledger. Any later write to a reviewed artifact
 rotates `artifactRevisionId` and makes this receipt stale — that is the contract
-working, not a mistake.
+working, not a mistake. When a missing or stale receipt is the reason for re-entry,
+the re-entry command is the **emitted** `next.continuation` (`rendering` for display):
+quote it, never author exact command tokens.
 
 One `PreExecutionReviewReceipt v1`
 (`agentic-workflow/pre-execution-review-receipt@1`) per review, in a fenced block:

package/skills/workflow-status/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: workflow-status
 user-invocable: true
-version: 3.6.0
+version: 3.7.0
 author: "Gabriel Trabanco <gtrabanco@users.noreply.github.com>"
 license: MIT
 argument-hint: "[--json-only] [--last-envelope <json|path>]"
@@ -117,6 +117,9 @@ the JSON contract.
 - The human summary (unless `--json-only`) and the envelope are printed, envelope
   last.
 - Nothing was modified anywhere.
+- When the envelope carries `next.continuation`, the next-command echo **quotes the
+  emitted `next.continuation`** (`rendering` for display), never author exact command
+  tokens; the human-facing prose `→ Next:` block below stays.
 
 → Next: the envelope's `next.recommended` command — it is computed from the
   actual state, so it IS the recommendation

package/skills/workflow-status/references/PRE_EXECUTION.md

@@ -41,7 +41,7 @@ sensor reports a missing gate, never a verdict that no review ever happened.
 |---|---|---|
 | `current` | stage PASS verdict **and** recomputed digest = bound digest | spec: `/plan-feature <slug>` · plan: `/execute-phase <NN>` |
 | `missing` | no receipt block for the stage | `defined` → `/review-spec <slug>` · planned/in-progress → `/review-plan <NN>` |
-| `stale` | bound digest differs, or a bound context row moved | re-run **that stage's** review |
+| `stale` | bound digest differs, or a bound context row moved | re-run **that stage's** review — quote the emitted planning-gate re-run continuation (`next.continuation`, `rendering` for display), never author exact tokens |
 | `wrong-stage` | the only PASS belongs to the other stage | the stage that is missing — never the stage that passed |
 | `substitute` | prose verdict, legacy block, or internal attestation stands where a receipt should be | that stage's review, and report the substitute as a blocker detail |
 | `self-approved` | the author of the artifact recorded the PASS | that stage's review in a clean context |
@@ -50,9 +50,11 @@ sensor reports a missing gate, never a verdict that no review ever happened.
 | `impossible-timeline` | `structural.fresh: false`, `structural.reasonCode: impossible-timeline` — digest and revision match by construction; a receipt's own recorded timeline is physically impossible (finish predating source revision's commit date beyond the published skew); a legacy receipt with no parsable Started/finished: line, or a source revision git cannot resolve, stays unflagged (fail-open) | that stage's review — the timeline is self-refuting, not a stale source |
 
 A stale receipt re-runs the **review**, not the authoring skill: the artifacts may be
-exactly right and simply need re-judging in a clean context. A missing verdict block,
-or a verdict outside the stage's fixed set, is `missing` with the reason recorded —
-never a guess in the recommended direction.
+exactly right and simply need re-judging in a clean context. The command to re-run is
+the **emitted** `next.continuation` (`rendering` for display) — quoted as-is, never
+authored as fresh command prose. A missing verdict block, or a verdict outside the
+stage's fixed set, is `missing` with the reason recorded — never a guess in the
+recommended direction.
 
 ### Envelope projection
 

package/README.es.md

@@ -1,196 +0,0 @@
-# @gtrabanco/pi-agentic-workflow
-
-> 🇬🇧 [English version](README.md)
-
-Una sola instalación del método
-[agentic-workflow](https://github.com/gtrabanco/agentic-workflow) en
-[Pi](https://github.com/badlogic/pi-mono): las skills canónicas, un comando por
-cada una y, si quieres, modelo distinto por comando que devuelve tu sesión
-después.
-
-- **Skills canónicas, sin cambios.** El paquete trae los mismos archivos
-  `SKILL.md` de este repositorio, byte a byte — sin bifurcación para Pi que se
-  desincronice.
-- **Comandos claros.** Escribes `/plan-feature --next`, no
-  `/skill:plan-feature --next`.
-- **Enrutamiento que puedes olvidar.** Por defecto no se configura nada: cada
-  comando corre en el modelo que ya tenías.
-
-## Instalación
-
-```sh
-pi install npm:@gtrabanco/pi-agentic-workflow
-```
-
-Reinicia Pi. `/agentic-workflow-settings` y los comandos de workflow quedan
-disponibles en cualquier proyecto. Si antes copiaste las skills a mano a
-`~/.pi/agent/skills`, borra esa copia: el paquete ya las aporta, y dos copias
-significan dos versiones del mismo método.
-
-## Comandos
-
-Cada skill incluida cuyo frontmatter dice `user-invocable: true` obtiene un
-comando con el mismo nombre. La lista se lee de las skills al arrancar, así que
-añadir una skill añade su comando — no existe una tabla de alias que mantener.
-Las skills internas que componen una skill pública (los pasos de revisión, el
-preflight de planificación, el contrato del envelope) viajan en el paquete pero
-no reciben comando propio — los componen los de arriba:
-
-| Comando | Para qué lo usas |
-| --- | --- |
-| `/audit-docs` | Comprobar que docs, roadmap, código y el índice de fixes coinciden. |
-| `/audit-pr` | La puerta de merge: ¿está este PR listo? |
-| `/design-feature` | Convertir una idea suelta en un SPEC diseñado. |
-| `/discover-repository-state` | Congelar hechos verificados del repositorio. |
-| `/execute-phase` | Implementar las fases restantes de una unidad planificada. |
-| `/fold-findings` | Reparar los hallazgos «fix-now» persistidos. |
-| `/generate-docs` | Generar guías incrementales basadas en el diff. |
-| `/init-workspace` | Adaptar el andamiaje del workflow a un repositorio. |
-| `/log-session` | Añadir una entrada de sesión estructurada a `docs/LOGS.md`. |
-| `/plan-feature` | Dirigir el trabajo diseñado a planificación y roadmap. |
-| `/plan-fix` | Redactar un SPEC de fix por fases desde uno o varios issues. |
-| `/product-audit` | Auditar la superficie del producto, no solo el diff. |
-| `/resolve-repository-state` | Resolver una contradicción en hechos congelados. |
-| `/review-change` | Revisar un cambio con los ejes que apliquen. |
-| `/review-plan` | Revisar un plan congelado en un contexto limpio. |
-| `/review-spec` | Revisar un SPEC diseñado en un contexto limpio. |
-| `/ship-roadmap` | Encontrar o continuar un roadmap, una etapa por ejecución. |
-| `/triage-issue` | Verificar un issue o hallazgo contra el código actual. |
-| `/workflow-status` | Estado de solo lectura del repositorio y el roadmap. |
-
-Los argumentos se reenvían tal cual: `/execute-phase P3 --fix` llega a la skill
-como `P3 --fix`.
-
-## Enrutamiento por modelo
-
-Dos archivos JSON, ambos opcionales:
-
-| Alcance | Ruta | Se lee cuando |
-| --- | --- | --- |
-| Global | `~/.pi/agent/pi-agentic-workflow.json` | siempre |
-| Proyecto | `<repo>/.pi/pi-agentic-workflow.json` | el proyecto es de confianza |
-
-```json
-{
-  "default": { "model": "anthropic/claude-opus-4-5", "thinking": "high" },
-  "commands": {
-    "plan-feature": { "model": ["anthropic/claude-sonnet-4-5", "openai/gpt-5.2"], "thinking": "medium" },
-    "review-change": { "thinking": "max" }
-  },
-  "onUnavailableRoute": "stop",
-  "onSettle": "keep"
-}
-```
-
-Cada valor se toma del primer sitio que lo declara: **comando en proyecto →
-comando global → ruta por defecto resuelta → defecto del paquete**. `review-change`
-de arriba usa el modelo del default con thinking `max`; todo lo demás corre con lo
-que ya tuviera la sesión, porque la ruta por defecto del paquete es
-`{"model": "inherit", "thinking": "inherit"}`.
-
-- `model` puede ser `provider/modelId` — la referencia exacta que muestra
-  `/model` — `"inherit"`, o un **array** ordenado de 1–4 referencias (una cadena
-  de respaldo). Para una cadena, el dispatch prueba cada referencia en orden y
-  aplica la primera que se resuelve y tiene credenciales, sin tocar la sesión
-  mientras prueba; cuando todas las entradas son inservibles el comando se
-  detiene (o, con `onUnavailableRoute` en `inherit`, avisa y corre con el modelo
-  actual), nombrando cada candidato y por qué se saltó. `plan-feature` de arriba
-  prueba `anthropic/claude-sonnet-4-5` primero y luego `openai/gpt-5.2`.
-- `thinking` es uno de `off`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max`,
-  o `"inherit"`.
-- `` `onSettle` `` es uno de `"keep"` (el default) o `"restore"` — ver
-  [Tu sesión después de un comando](#tu-sesión-después-de-un-comando).
-- Claves desconocidas, `null` y referencias mal formadas se **rechazan**, no se
-  ignoran: un error tipográfico que no hace nada en silencio es el bug que nunca
-  encuentras.
-
-El primer comando de workflow tras la instalación avisa una sola vez que el
-enrutamiento es configurable, y no vuelve a insistir. Ese aviso se guarda en
-`~/.pi/agent/pi-agentic-workflow-state.json`, no en tu configuración.
-
-## Cuando el modelo configurado no está disponible
-
-Por defecto el comando **se niega a arrancar** y te dice por qué: el modelo no
-está en el registro, no tiene credenciales, o no se pudo seleccionar. No se envía
-nada, así que nada corre en un modelo que no elegiste. Para usar el modelo actual
-aun así:
-
-```json
-{ "onUnavailableRoute": "inherit" }
-```
-
-## Tu sesión después de un comando
-
-Por defecto (**`onSettle: "keep"`**), el modelo enrutado y el nivel de thinking
-**se quedan** en la ventana de chat abierta cuando el comando termina. Si
-`plan-feature` corre en `glm-5` y quieres retocar el plan que produjo, tu siguiente
-prompt sigue corriendo en `glm-5` — igual con una pregunta posterior. Cuando no
-necesitas el modelo pesado, lo cambias tú con `/model` (o Ctrl+P) a algo más
-barato; nada restaura sobre tu elección.
-
-Para volver a traer el modelo y el nivel de thinking previos al comando cuando este
-termina, pon:
-
-```json
-{ "onSettle": "restore" }
-```
-
-(El modo `restore` es el histórico contrato AC8: tras un comando enrutado la
-sesión vuelve a como estaba — el modelo *y* el nivel de thinking, porque
-seleccionar un modelo puede mover el nivel. Si cambias el modelo tú mismo a
-mitad de turno, con `/model` por ejemplo, no se restaura nada: tu elección gana,
-y el comando lo dice. Si mueves solo el nivel de thinking, lo conservas mientras
-el modelo vuelve.)
-
-## Consola de configuración
-
-```
-/agentic-workflow-settings
-```
-
-`/aw-settings` es un atajo para la misma consola.
-
-Muestra en qué corre cada comando ahora mismo, y qué archivo se niega a parsearse,
-y luego deja editar **un archivo a la vez** y guardar en alcance global o de
-proyecto. No guarda encima de un archivo que no sabe leer, y no toca el
-archivo de proyecto mientras el proyecto no sea de confianza.
-
-Los selectores de modelo y thinking de la consola son buscables y con ventana:
-escribir estrecha la lista, el cursor se queda en pantalla, un indicador de
-posición muestra dónde estás y el valor en vigor viene preseleccionado y
-etiquetado `(current)` / `(default route)`. Una edición de ruta pregunta qué
-campo cambiar (model, thinking), así que guardar sin cambios deja el archivo
-byte a byte idéntico; el campo de modelo puede construir una cadena de respaldo
-ordenada (`a/m1 → b/m2`); y una sola pasada puede aplicar o limpiar una ruta en
-varios comandos, avisando por comando cuando una referencia falta del registro
-vivo. Fuera de una sesión TUI los selectores caen a un prompt simple, así que la
-consola nunca se queda sin salida. Las listas de más de 24 opciones tampoco
-llegan nunca a un solo diálogo: por encima del tope el selector de modelo
-pregunta primero el proveedor, y una lista de proveedores o de modelos
-demasiado larga pagina (`More options…` / `◀ Previous page`) en vez de tumbar
-el diálogo de select del host.
-
-## Diagnóstico
-
-| Ves | Significa |
-| --- | --- |
-| `refused: invalid configuration` | Un archivo de configuración fue rechazado. El mismo mensaje nombra el campo, p. ej. `$.commands.plan-feature.model`. Ejecuta `/agentic-workflow-settings` para ver el archivo o arregla el JSON. |
-| `stopped: the configured model` … `is not in the model registry` | La referencia es incorrecta o el proveedor no está configurado. Usa `/model` para ver el `provider/modelId` exacto. |
-| `has no configured credentials` | El modelo existe pero aún no puedes usarlo. Autentica, o pon `onUnavailableRoute` en `inherit`. |
-| `could not be selected` | Pi rechazó el cambio. El comando se detiene con el motivo — salvo que `onUnavailableRoute` sea `inherit`, en cuyo caso avisa y se ejecuta con tu modelo actual. |
-| `refused: the agent is busy` | Hay un turno en ejecución. Espera a que termine. |
-| `is still routed` | El comando enrutado anterior no ha terminado. |
-| `leaving the model you chose in place` | (en modo `restore`) Cambiaste el modelo durante un turno enrutado, así que no se restauró nada: tu elección ganó. |
-| `these configured routes match no command` | Una clave de `commands` no nombra nada. Corrige la escritura o elimina la entrada. |
-
-## Notas
-
-- Probado con Pi 0.85.1 (2026-09-05) (`pi install`, skills de paquete, registro de
-  comandos amigables, set/clear enrutado, round-trip de consola de ajustes,
-  `sendUserMessage` con expansión de plantillas).
-- El paquete declara Pi como peer dependency; no incluye ninguna copia de Pi.
-- Las skills pueden indicar al modelo que ejecute comandos. Revísalas como con
-  cualquier paquete de terceros.
-
-MIT · [Repositorio](https://github.com/gtrabanco/agentic-workflow) ·
-[`docs/features/27-pi-agentic-workflow/`](../../docs/features/27-pi-agentic-workflow/SPEC.md)