@gtrabanco/pi-agentic-workflow 0.34.2 → 0.35.1

package/README.md

@@ -146,16 +146,27 @@ set of routes, and you can define as many as you like under `profiles`:
   The built-in `nan` profile is shown but not editable — create a profile to
   override it.
 
-## Serena (ships by default)
+## Serena (native MCP registration)
 
-Installing this package also gives you **Serena's language-server tools** —
-`serena_find_symbol`, `serena_find_referencing_symbols`,
-`serena_get_symbols_overview`, `serena_find_declaration`, diagnostics and the
-rest. Two declarations make that work, and they ship together: the exact-pinned
-`@bacnh85/pi-serena` dependency, and a second `pi.extensions` entry pointing at
-that dependency's own extension. Pi installs a package's dependencies but
-activates only the package's own `pi.extensions` entries, so the pin alone would
-install a toolset nothing ever loaded.
+Serena's language-server tools reach pi through Serena's own MCP server — the
+interface Serena documents, tests and keeps version-stable — not through a
+bundled bridge. This package ships no Serena dependency; provisioning is one
+consented step:
+
+```sh
+pi mcp add serena -- serena start-mcp-server --context claude-code
+pi mcp list   # verifies the connection
+```
+
+That writes the user-level `~/.pi/agent/mcp.json` (add `--local` for a
+project-level `.pi/mcp.json`, read only after project trust is granted). The
+tools appear under the `mcp__serena__<tool>` namespace — `find_symbol`,
+`find_referencing_symbols`, `get_symbols_overview`, `get_diagnostics` and the
+rest — and pi's per-server `toolExposure` config narrows the surface if the
+full toolset is more than a session needs. The machine half is the same local
+tool as before: `uv tool install -p 3.13 serena-agent` (or `/init-workspace`,
+which offers the install, the registration and the global trust append in one
+consented question).
 
 **You must run `/init-workspace` once per project.** This package never writes or
 repairs `.serena/project.yml` — `init-workspace` is its only writer, so there is
@@ -165,27 +176,12 @@ missing, lists no `language_servers`, or was provisioned by an older bundled
 `/init-workspace`; the latch lives in this package's state file, so a restart does
 not re-nag.
 
-Serena itself is a local tool: install the machine half with `uv tool install -p
-3.13 serena-agent` (or `/init-workspace`, which offers it). Nothing is uploaded
-and no key is configured.
-
-**Switching it off.** pi owns package resources, so the off switch is pi's, not a
-key of ours: `pi config` opens the resource picker (tab switches scope) and lets
-you disable the package's resources, and the same filter is expressible in
-settings under `packages`:
-
-```json
-{
-  "packages": [
-    { "source": "npm:@gtrabanco/pi-agentic-workflow", "extensions": ["./dist/extension/index.js"] }
-  ]
-}
-```
+Nothing is uploaded and no key is configured.
 
-That keeps the workflow's own extension and commands and drops the Serena
-extension entry, so the `serena_*` tools are never registered. There is no
-settings key inside this package for it: the entry is declared in the manifest,
-and a config key cannot gate a manifest-declared extension.
+**Switching it off.** Serena is an MCP server now, so pi's own MCP controls are
+the off switch: `pi mcp remove serena` unregisters it, `/mcp` inside a session
+disables or re-enables it, or set `"enabled": false` on the entry in
+`~/.pi/agent/mcp.json` to keep it listed but disconnected.
 
 ## Path protection (optional)
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gtrabanco/pi-agentic-workflow",
-  "version": "0.34.2",
+  "version": "0.35.1",
   "description": "Pi package: canonical agentic-workflow skills, friendly slash commands, and per-command model routing.",
   "keywords": [
     "pi-package",
@@ -21,8 +21,7 @@
   "type": "module",
   "pi": {
     "extensions": [
-      "./dist/extension/index.js",
-      "./node_modules/@bacnh85/pi-serena/extensions/index.ts"
+      "./dist/extension/index.js"
     ],
     "skills": [
       "./skills"
@@ -58,7 +57,6 @@
     "@earendil-works/pi-tui": "*"
   },
   "dependencies": {
-    "@bacnh85/pi-serena": "0.10.0",
     "@gtrabanco/agentic-workflow": "0.4.0",
     "@gtrabanco/agentic-workflow-schema": "4.8.0"
   },

package/skills/execute-phase/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: execute-phase
 user-invocable: true
-version: 6.2.2
+version: 6.3.1
 argument-hint: <NN-slug> [P<k>|step] [--max-attempts N] [--force]
 allowed-tools: [Bash, Read, Edit, Write, MultiEdit]
 author: "Gabriel Trabanco <1969593+gtrabanco@users.noreply.github.com>"
@@ -182,7 +182,13 @@ lane hand-off is a defect; re-read the hand-off before looping. After each
 step: gates, execute, diff guard (if implement), evidence/progress update,
 commit. Skip intermediate review checkpoints; the mandatory end review covers
 the full accumulated diff. On gate red at any step: repair that step (up to
-`--max-attempts`), then continue the loop. Terminal:
+`--max-attempts`), then continue the loop. The close-out is part of the last
+step's completion: open the PR on `main` with `Closes #N`, write `- PR: <url>`
+into References, and settle the index row with `agentic-workflow roadmap
+<index> rowUpsert '<json>'` (`docs/features/ROADMAP.md`; a fix unit edits its
+`docs/fix/README.md` status cell — no `NN` column, no `rowUpsert`).
+`bun scripts/roadmap-settle.mjs` exits 0 (#316).
+Terminal:
 
 ```text
 UNIT LOOP — <unit> COMPLETE

package/skills/init-workspace/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: init-workspace
 user-invocable: true
-version: 4.2.0
+version: 4.3.0
 argument-hint: <target-dir>
 author: "Gabriel Trabanco <1969593+gtrabanco@users.noreply.github.com>"
 license: MIT

package/skills/init-workspace/references/SERENA.md

@@ -172,6 +172,29 @@ entry, never another line.
 | Cursor | `.cursor/mcp.json` → `mcpServers.serena` | `{"command":"serena","args":["start-mcp-server","--context=ide"]}` |
 | JetBrains Copilot | the global `mcp.json` Copilot opens | `{"type":"stdio","command":"serena","args":["start-mcp-server","--context=jb-copilot-plugin"]}` |
 | Codex | `~/.codex/config.toml` → `[mcp_servers.serena]` | `command = "serena"`, `args = ["start-mcp-server","--project-from-cwd","--context=codex"]` |
+| **pi** (this agent harness) | `pi mcp add serena -- serena start-mcp-server --context claude-code` (user-level default → `~/.pi/agent/mcp.json`; project-level: add `--local` → `.pi/mcp.json`) | command `serena` |
+
+**pi-specific notes:** the native MCP registration replaces the old `@bacnh85/pi-serena` bridge (removed by issue #321). Two options:
+
+- **User-level (default)** — `pi mcp add serena -- serena start-mcp-server --context claude-code` writes `~/.pi/agent/mcp.json`; fresh clones are zero-config. Ask first, like every other global write.
+- **Project-level** — `pi mcp add --local serena -- serena start-mcp-server --context claude-code` writes `.pi/mcp.json` (read only after project trust is granted), keeping the choice in git:
+  ```json
+  {
+    "mcpServers": {
+      "serena": {
+        "command": "serena",
+        "args": ["start-mcp-server", "--context", "claude-code"]
+      }
+    }
+  }
+  ```
+  Same consent rule as the other per-workspace files (`.vscode/mcp.json`, `.cursor/mcp.json`).
+
+Verify either option with `pi mcp list` (it connects and exits 1 on a broken entry). Tools are namespaced `mcp__serena__<tool>` (e.g. `mcp__serena__find_symbol`), so generic native names cannot collide with other MCP servers. If the full tool surface is too wide for the session, pi's per-server `toolExposure` config narrows it — the same knob the bridge solved by shipping a curated 22.
+
+The `--context claude-code` value resolves issue #321's open question: pi is a terminal agent harness like Claude Code, so it uses Serena's `claude-code` context (not the REPL `agent_interface`, which stays out of scope — the classic `tools` interface remains the default).
+
+**Consent gating:** the registration is asked inside the same consented question as the tool install and trust append. A project-level `.pi/mcp.json` write additionally commits a file to the repo, so it needs the same explicit yes as any other write.
 
 ### Trust — the fresh clone (global config, consent-gated)
 

package/skills/unit-lane/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: unit-lane
 user-invocable: true
-version: 2.4.3
+version: 2.5.1
 argument-hint: <NN-slug | "<idea>"> [--retriage] | --fix <issue-number> | --from-issue <issue-number>
 description: >
   One-shot lane conductor for a delivery unit: triage, implement, evidence, and

package/skills/unit-lane/references/PLAN.md

@@ -100,6 +100,11 @@ P1 — <task description> (validator: <backticked command> <expected outcome>)
 P2 — <task description> (validator: <backticked command> <expected outcome>)
 …
 Pn — last task is verification when the unit has behavior
+P<n+1> — release settle, when the unit opens a PR: after the PR exists, settle
+  the roadmap row via the crate's writer (`agentic-workflow roadmap rowUpsert`,
+  status `done · [#N](<pr-url>)`, every other cell preserved) — validator:
+  `bun scripts/roadmap-settle.mjs` exits 0 (#316: the settle is a task the
+  executor owns, never a missing courtesy)
 ```
 
 Both phase grammars are accepted (owner: `phase-contract`): these bullets, or
@@ -164,7 +169,8 @@ name those files instead.
 - [ ] Every AC maps to at least one task, and no task re-decides an AC
 - [ ] Tasks are ordered smallest first
 - [ ] Each task names its validator (a backticked command plus its expected outcome)
-- [ ] Final task is verification (proves the whole unit works)
+- [ ] Final task is verification (proves the whole unit works); a unit that
+      opens a PR ends with the release-settle task (#316)
 - [ ] No task reaches into a future-phase concern (docs, release, migrations)
 - [ ] Phases labeled P1, P2, … — `#308`: `S#` is the requirement id (a spec),
       `P#` is the phase label (a task); the two vocabularies never swap roles
@@ -190,7 +196,8 @@ name those files instead.
 - Do not decide at the ask stage, and do not re-ask an answered question
 - Do not bundle across phase boundaries — one phase = one commit
 - Do not plan work beyond the unit's acceptance criteria
-- Do not include docs/release tasks here — those are separate catalog steps
+- Do not include docs/release tasks here — those are separate catalog steps;
+  the release-settle task above is the sole exception (#316)
 - Do not invent tasks that the ACs do not demand
 - Do not turn the sweep into a gate, stage verdict or blocking check — it routes,
   it never blocks a miss, and it adds no authority

package/skills/unit-lane/references/RELEASE.md

@@ -30,7 +30,13 @@ Write into the unit doc's **References** section:
 - [ ] Branch follows the project's Workflow conventions
 - [ ] Commit messages follow conventional format (`type(scope): description`)
 - [ ] PR body includes the Evidence section as verification record
-- [ ] Roadmap row updated to `done` (or `merged` per conventions)
+- [ ] Index row settled via the crate's writer — feature: re-emit the row with
+      `agentic-workflow roadmap docs/features/ROADMAP.md rowUpsert
+      '{"nn":"<NN>","slug":"<slug>","status":"done · [#<pr>](<pr-url>)"}'`,
+      every other cell as it is; fix: the `docs/fix/README.md` status cell — that
+      index has no `NN` column, so `rowUpsert` (the roadmap's writer) does not
+      apply to it. `bun scripts/roadmap-settle.mjs` exits 0 afterwards (issue
+      #316: the settle is code, never prose)
 - [ ] `git status --porcelain` is empty
 
 ## Allowed