@zosmaai/pi-llm-wiki 0.9.0 → 0.9.2

package/CHANGELOG.md

@@ -3,9 +3,16 @@
 ## [Unreleased]
 
 ### Fixed
+- **`wiki_watch` referred to a fictional `schedule_prompt` tool** (Issue #81): `wiki_watch` emitted instructions to run `schedule_prompt action=add ...` — a tool that doesn't exist in pi-coding-agent or any pi extension — so `/wiki-run --schedule weekly` printed a command the user couldn't execute. The emitted payload also contained the typo `/wiki:run` (real slash-command is `/wiki-run`), and the tool description claimed it scheduled jobs when it only ever printed instructions. Now emits a real, copy-pasteable **POSIX 5-field crontab line** wrapped in `/bin/bash -lc` (so npm-global / bun / nvm PATH is imported under cron's minimal env), with defensive `mkdir -p` of the log dir and a `# llm-wiki-autoupdate` removal tag. The tool description, output text, `prompts/wiki-run.md`, `docs/api.md` and all 10 i18n READMEs now explicitly say "prints — does not install". `details` now carries `cronLine` and `installed: false`. New `test/wiki-watch.test.ts` (10 tests) locks down all three regressions plus the portability contract (login-shell wrapper, defensive `mkdir`, quoted `$HOME`).
 - **Personal wiki created at doubled path `~/.llm-wiki/.llm-wiki/…`**: `getPersonalWikiRoot()` returned the dot-dir itself (`~/.llm-wiki`) while `getVaultPaths()` then appended another `.llm-wiki/` segment, so the personal vault was written to `~/.llm-wiki/.llm-wiki/wiki/…`. Fixed by aligning `getPersonalWikiRoot()` with the same "root = parent of `.llm-wiki/`" contract used by project vaults. `WIKI_HOME` continues to override the parent.
 
 ### Added
+- **Visible wiki activity + background/reported mutations** (Issue #77): the wiki was effectively invisible — recall was appended only to the **system prompt**, the observe/retro reminder was sent with `display: false`, and the lone user-facing cue was a static status line. The wiki now surfaces what it does, and mutating work is pushed off the agent's critical path.
+  - **Visible surfaces**: a one-time **session notice** (`buildSessionNotice`) announces the full loop — retrieval (recall → `wiki_search` → `read`, all synchronous because the LLM consumes their output) and capture (`wiki_observe` → `wiki_retro`, background + reported); the periodic reminder is now `display: true` and names **both** capture tools (`buildReminderText`); and the status line becomes **recall-aware** (`🧠 LLM Wiki — recalled N page(s) for this task`) when auto-recall matches.
+  - **Background + reported principle**: only `wiki_search` / `read` / `wiki_recall` stay synchronous. Heavy mutations — `wiki_rebuild_meta`, `wiki_reindex_embeddings`, `wiki_lint` — now dispatch to the background runtime and **report a visible completion message** instead of blocking the turn; `wiki_ingest` gained a persistent completion report alongside its toast. New `Runtime.report()` / `Runtime.launchReported()` primitives and a `dispatchReported()` tool helper (with a synchronous fallback when no runtime is available, preserving prior behavior and unit tests).
+  - **`notices` setting** (namespaced `llm-wiki`, default **on**): set `false` to restore the previous quiet behavior — static status line and silently-injected (`display: false`) reminders/reports.
+  - **Fixed** a dangling reference to a non-existent `wiki_read` tool in the links-first recall output (now points at `read`).
+  - **16 tests** (`test/visible-activity.test.ts`, `test/background-tools.test.ts`): `notices` parsing/defaulting, reminder + session-notice content, `Runtime.report`/`launchReported` (display gating, no-pi no-op, error isolation, null-summary skip), and `wiki_rebuild_meta` background-dispatch + report vs synchronous fallback.
 - **Model selection surface for background tasks** (Issue #69, part of epic #63): the wiki background lane (ingest synthesis, etc.) now has a user-facing surface to choose its model, defaulting to the **session model** with zero config.
   - **`/wiki-model` slash command**: run with no argument for an interactive picker (lists `modelRegistry.getAvailable()`); `/wiki-model provider/id` to set directly (scriptable, no UI needed); `/wiki-model session` (or `clear`/`default`/`reset`) to revert to the session model. The choice is **persisted** to project settings (`.pi/settings.json` under `llm-wiki.taskModel`, preserving other keys) and applied immediately, with a status-bar label of the active model.
   - **Per-call `model` override** on heavy tools (`wiki_ingest`): an optional `'provider/id'` param that overrides the configured `taskModel` for that one call. Precedence is **override > configured taskModel > session model**; each layer is applied only when the model is in the registry, and a missing/unknown layer warns (when UI is available) and falls through — so a bad ref degrades gracefully instead of failing.

package/README.de.md

@@ -78,7 +78,7 @@ Das Ergebnis ist ein Wiki, das wächst, während du Quellen erfasst, Fragen stel
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ Das Ergebnis ist ein Wiki, das wächst, während du Quellen erfasst, Fragen stel
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### Schrägstrich-Befehle
 

package/README.es.md

@@ -78,7 +78,7 @@ El resultado es un wiki que se acumula a medida que capturas fuentes, haces preg
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ El resultado es un wiki que se acumula a medida que capturas fuentes, haces preg
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### Comandos de Barra
 

package/README.fr.md

@@ -78,7 +78,7 @@ Le résultat est un wiki qui s'accumule au fur et à mesure que vous capturez de
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ Le résultat est un wiki qui s'accumule au fur et à mesure que vous capturez de
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### Commandes Slash
 

package/README.hi.md

@@ -78,7 +78,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### स्लैश कमांड
 

package/README.ja.md

@@ -78,7 +78,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### スラッシュコマンド
 

package/README.ko.md

@@ -78,7 +78,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### 슬래시 명령
 

package/README.md

@@ -78,10 +78,11 @@ The result is a wiki that **compounds** as you capture sources, ask questions, a
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
+| 🧭 **Agent working-memory** _(opt-in)_ | `wiki_capture_trajectory` records *how* a task was solved (tool-call trajectory) → distill into reusable `skill`/`case` pages → `wiki_recall_skill` surfaces them next time. Off by default; enable with `/wiki-trajectories on` |
 | 🌐 **MCP Server** | Use with Claude Code, Cursor, Windsurf via stdio MCP transport |
 | 📝 **Obsidian-friendly** | Folder-qualified wikilinks, stable source-ID citations, compatible vault |
 | 🛡️ **Guardrails** | Blocks direct edits to raw sources and generated metadata |
@@ -105,7 +106,12 @@ The result is a wiki that **compounds** as you capture sources, ask questions, a
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
+| `wiki_capture_trajectory` _(opt-in)_ | Capture the completed task's tool-call trajectory (agent working-memory) |
+| `wiki_distill_skills` _(opt-in)_ | Batch undistilled trajectories for synthesis into reusable skill pages |
+| `wiki_recall_skill` _(opt-in)_ | Recall distilled skills + similar past cases — "have I done this before?" |
+
+> The three agent-trajectory tools are **off by default** (issue #80). Enable them with `/wiki-trajectories on` (sets `llm-wiki.trajectories`); when off they are not registered at all.
 
 ### Slash Commands
 
@@ -121,6 +127,9 @@ The result is a wiki that **compounds** as you capture sources, ask questions, a
 | `/wiki-digest [--period daily\|weekly]` | Generate a digest of recent activity |
 | `/wiki-retro` | Save atomic insights from completed tasks |
 | `/wiki-req <concept>` | Decompose a concept into atomic, traceable requirement pages |
+| `/wiki-trajectories <on\|off>` | Enable/disable agent working-memory (opt-in, off by default) |
+| `/wiki-record <title>` | Capture the completed task's trajectory (requires trajectories enabled) |
+| `/wiki-skills [query]` | Search distilled skills + past cases (requires trajectories enabled) |
 
 ---
 
@@ -425,6 +434,20 @@ Thanks to everyone who has contributed! This list is regenerated automatically b
                     <br />
                     <sub><b>Shanvit S Shetty</b></sub>
                 </a>
+            </td>
+            <td align="center">
+                <a href="https://github.com/CelestialCreator">
+                    <img src="https://avatars.githubusercontent.com/u/177931942?v=4" width="64;" alt="CelestialCreator"/>
+                    <br />
+                    <sub><b>Akshay</b></sub>
+                </a>
+            </td>
+            <td align="center">
+                <a href="https://github.com/mystery4f">
+                    <img src="https://avatars.githubusercontent.com/u/40482524?v=4" width="64;" alt="mystery4f"/>
+                    <br />
+                    <sub><b>标准萌新</b></sub>
+                </a>
             </td>
 		</tr>
 	<tbody>

package/README.pt.md

@@ -78,7 +78,7 @@ O resultado é um wiki que se acumula conforme você captura fontes, faz pergunt
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ O resultado é um wiki que se acumula conforme você captura fontes, faz pergunt
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### Comandos de Barra
 

package/README.ru.md

@@ -78,7 +78,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### Слэш-команды
 

package/README.zh.md

@@ -78,7 +78,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | 🔍 **Full-text search** | Generated registry with keyword lookup across all pages and sources |
 | 🩺 **Mechanical linting** | Orphans, broken links, duplicate aliases, coverage gaps, stale captures |
 | 📊 **Dashboard** | `wiki_status` — counts, source states, recent activity |
-| 🤖 **Auto-update watch** | `wiki_watch` — schedule periodic discovery + ingest |
+| 🤖 **Auto-update watch** | `wiki_watch` — print a `crontab` line that runs the full cycle on a schedule |
 | 🧠 **Layered recall** | Searches both personal (`~/.llm-wiki/`) and project (`.llm-wiki/`) vaults — personal knowledge follows you everywhere |
 | 📝 **Auto-bootstrap** | Extension suggests creating a wiki when none exists in the current directory |
 | 💾 **Lightweight capture** | `wiki_retro` — save atomic insights as a single markdown file; full 4-layer pipeline also available via `wiki_capture_source` |
@@ -105,7 +105,7 @@ The extension will proactively suggest creating a wiki on your first session. Al
 | `wiki_status` | Show counts, source states, and recent activity |
 | `wiki_rebuild_meta` | Force a full metadata rebuild (registry, backlinks, index, log) |
 | `wiki_log_event` | Append a structured event to the wiki activity log |
-| `wiki_watch` | Schedule automatic wiki updates (daily / weekly / hourly) |
+| `wiki_watch` | Print a `crontab` line for automatic wiki updates (daily / weekly / hourly) — does not install it |
 
 ### 斜杠命令
 

package/docs/api.md

@@ -1,6 +1,11 @@
 # API Reference
 
-All 13 tools registered by the extension. Parameters marked `?` are optional.
+All tools registered by the extension. Parameters marked `?` are optional.
+
+13 tools are always registered. The 3 agent-trajectory tools
+(`wiki_capture_trajectory`, `wiki_distill_skills`, `wiki_recall_skill`) are **opt-in,
+off by default** (issue #80) — they are only registered when `llm-wiki.trajectories`
+is `true`; enable with `/wiki-trajectories on`.
 
 ---
 
@@ -92,7 +97,7 @@ Resolve or safely create a canonical wiki page. Returns immediately if the page
 
 | Name | Type | Required | Description |
 |------|------|----------|-------------|
-| `type` | `string` | ✅ | Page type: `"entity"`, `"concept"`, `"synthesis"`, `"analysis"`, or `"requirement"` |
+| `type` | `string` | ✅ | Page type: `"entity"`, `"concept"`, `"synthesis"`, `"analysis"`, `"requirement"`, `"skill"`, or `"case"` |
 | `title` | `string` | ✅ | Human-readable page title; auto-slugified to a kebab-case filename |
 | `content` | `string` | — | Full markdown content for the page; if omitted, the type-appropriate template is used |
 
@@ -309,28 +314,136 @@ details: { kind: string }
 
 ## wiki_watch
 
-Output the shell command needed to schedule automatic wiki updates (discover → ingest → lint) via
-pi's `schedule_prompt` cron system. Does not schedule anything directly — it returns the command
-for the user to run.
+Print a ready-to-paste **POSIX crontab line** that runs the full wiki cycle (discover → ingest →
+lint) on a schedule by invoking `pi -p "/wiki-run"` headlessly under `/bin/bash -lc` so the
+user's shell profile (and the `pi` binary on npm-global / bun / nvm PATH) is imported. **Does
+not schedule anything directly** — it returns the command for the user to install with
+`crontab -e`. Calling agents should surface the output verbatim and avoid claiming the schedule
+is active.
 
 **Parameters**
 
 | Name | Type | Required | Description |
 |------|------|----------|-------------|
-| `interval` | `string` | ✅ | `"daily"` (8:00 AM), `"weekly"` (Monday 9:00 AM), `"hourly"`, or `"stop"` (prints removal instructions) |
+| `interval` | `string` | ✅ | `"daily"` (8:00 AM), `"weekly"` (Monday 9:00 AM), `"hourly"`, or `"stop"` (prints crontab removal instructions) |
 
 **Returns**
 
 ```
 details: {
   interval: string,
-  cronSchedule: string,   // e.g. "0 0 8 * * *"
-  label: string           // e.g. "Daily at 8:00 AM"
+  cronSchedule: string,   // 5-field POSIX expression, e.g. "0 8 * * *"
+  label: string,          // e.g. "Daily at 8:00 AM"
+  cronLine: string,       // full crontab line, tagged "# llm-wiki-autoupdate"
+  installed: false        // tool never installs — always false
+}
+```
+
+Output is appended to `~/.llm-wiki/cron.log` (the directory is created by the cron line itself
+via `mkdir -p`). On systems without `/bin/bash`, replace the wrapper with `/bin/sh -c` and
+ensure `pi` is in cron's PATH yourself.
+
+When `interval` is `"stop"`, returns `details: { action: "stop_instructions" }` with
+instructions for removing the line via `crontab -e` (look for the `# llm-wiki-autoupdate` tag).
+
+---
+
+## wiki_capture_trajectory
+
+Capture the just-completed task's tool-call trajectory into an immutable packet
+(`raw/trajectories/TRJ-*`) with a self-contained summary (`extracted.md`). The working-memory
+counterpart to `wiki_capture_source`. By default the trajectory is auto-extracted from the live
+session; pass `steps` to override. **Opt-in** (issue #80): only available when
+`llm-wiki.trajectories` is enabled (`/wiki-trajectories on`).
+
+**Parameters**
+
+| Name | Type | Required | Description |
+|------|------|----------|-------------|
+| `title` | `string` | — | Short descriptive title for the task (≤60 chars). Inferred from the prompt if omitted. |
+| `task` | `string` | — | The task/prompt that started the work. Inferred from the session if omitted. |
+| `outcome` | `string` | — | `"success"` (default), `"failure"`, or `"partial"` — recorded in the packet manifest |
+| `steps` | `array` | — | Explicit trajectory steps (tool-call history). Omit to auto-extract from the live session. |
+| `model` | `string` | — | Model that ran the task. Inferred from the session if omitted. |
+
+**Returns**
+
+```
+details: {
+  trajectoryId: string,    // e.g. "TRJ-2026-06-07-001"
+  packetPath: string,      // path to raw/trajectories/TRJ-*/ (packet.json + extracted.md)
+  stepCount: number
+}
+```
+
+Errors with `isError: true` if no vault exists, or with `error: "empty_trajectory"` when no
+trajectory can be extracted and no `steps` are provided.
+
+---
+
+## wiki_distill_skills
+
+Return a batch of captured trajectories that have not yet been distilled into `skill` pages. Does
+not write anything itself — the model reads each packet and synthesizes reusable skill pages (via
+`wiki_ensure_page(type="skill")`) that cite the trajectory IDs. A trajectory counts as "distilled"
+once a `skills/` page links to it.
+
+**Parameters**
+
+| Name | Type | Required | Description |
+|------|------|----------|-------------|
+| `trajectory_id` | `string` | — | Distill a specific trajectory only; omit for all undistilled |
+| `batch_size` | `number` | — | Max trajectories to return (default: `3`, max: `5`) |
+
+**Returns**
+
+```
+details: {
+  batch: string[],    // trajectory IDs in this batch, e.g. ["TRJ-2026-06-07-001"]
+  remaining: number   // undistilled trajectories still waiting after this batch
+}
+```
+
+Each batch entry includes the title, step/tool-call counts, and paths to read
+(`raw/trajectories/{id}/packet.json` and `extracted.md`). Returns an "all trajectories distilled"
+message with `{ distilled, total }` when nothing is pending.
+
+---
+
+## wiki_recall_skill
+
+Search distilled `skill` pages and past `case` pages for patterns relevant to the current task —
+answers "have I done something like this before?". Filters layered recall (`searchWikiLayered`) to
+skill/case pages. Call at the START of a task.
+
+**Parameters**
+
+| Name | Type | Required | Description |
+|------|------|----------|-------------|
+| `query` | `string` | ✅ | Search query — use the task description or key terms |
+| `kind` | `string` | — | `"skill"`, `"case"`, or `"any"` (default) |
+| `max_results` | `number` | — | Maximum pages to return (default: `5`, max: `10`) |
+
+**Returns**
+
+```
+details: {
+  query: string,
+  kind: string,
+  matches: Array<{
+    id: string,           // folder-qualified page ID, e.g. "skills/jwt-revocation"
+    title: string,
+    type: string,         // "skill" | "case"
+    preview: string,
+    path: string,
+    score: number,
+    vaultLabel?: string   // "📓 personal" when result is from the personal vault
+  }>
 }
 ```
 
-When `interval` is `"stop"`, returns `details: { action: "stop_instructions" }` with instructions
-for removing existing jobs via `schedule_prompt action=remove`.
+Returns empty `matches: []` with a hint to capture work via `wiki_capture_trajectory` /
+`wiki_distill_skills` when nothing matches.
 
 ---
 

package/docs/architecture.md

@@ -43,12 +43,18 @@ WIKI_ROOT/
     │   ├── original/          # Original artifact
     │   ├── extracted.md       # Normalized markdown
     │   └── attachments/       # Downloaded images, PDFs
+    ├── raw/trajectories/TRJ-*/ # Immutable agent task packets (extension-owned)
+    │   ├── manifest.json      # Capture metadata (format: trajectory)
+    │   ├── packet.json        # Full tool-call sequence
+    │   └── extracted.md       # README summary
     ├── wiki/                  # Editable knowledge pages (you + LLM)
     │   ├── sources/           # One summary per source
     │   ├── entities/          # People, orgs, tools, products
     │   ├── concepts/          # Ideas, patterns, frameworks
     │   ├── syntheses/         # Cross-cutting analyses
-    │   └── analyses/          # Durable query answers
+    │   ├── analyses/          # Durable query answers
+    │   ├── cases/             # One specific past task per trajectory
+    │   └── skills/            # Reusable patterns distilled from trajectories
     ├── meta/                  # Auto-generated (extension-owned)
     │   ├── registry.json      # Master page catalog
     │   ├── backlinks.json     # Inbound link map
@@ -89,11 +95,42 @@ Each captured source becomes a packet:
 - **concept** — ideas, patterns, frameworks
 - **synthesis** — cross-source theses and tensions
 - **analysis** — durable filed answers from queries
+- **requirement** — atomic requirements with status, priority, and traceability
+- **trajectory** — an immutable captured agent task run (working-memory source)
+- **case** — one specific past task implementation, citing its trajectory
+- **skill** — a reusable pattern distilled from one or more trajectories
+
+## Agent Working-Memory (Trajectories)
+
+The wiki captures not only what the agent *reads* (sources) but what it *does*
+(trajectories). A completed task is just another kind of source, so it flows
+through the same pipeline:
+
+```
+raw/trajectories/TRJ-*  →  wiki/skills/*  (+ optional wiki/cases/*)  →  meta/*
+```
+
+This is **opt-in, off by default** (issue #80): the three tools below are only
+registered when `llm-wiki.trajectories` is enabled (`/wiki-trajectories on`), and
+the `raw/trajectories`, `wiki/skills`, `wiki/cases` directories are created lazily
+on first use — so a vault with the feature off carries no trace of it.
+
+- `wiki_capture_trajectory` writes the immutable packet + a self-contained summary
+  (`extracted.md`), auto-extracting the tool-call sequence from the live session.
+  It does not emit a to-be-fleshed skeleton — capture is a single lightweight call.
+- `wiki_distill_skills` batches undistilled trajectories so the model can
+  generalize them into reusable `skill` pages.
+- `wiki_recall_skill` filters layered recall to `skill`/`case` pages —
+  "have I done something like this before?".
+
+Trajectory packets live under `raw/**` and are therefore immutable under the
+same guardrail as source packets — no new ownership rule required.
 
 ## Linking Style
 
 - Internal: `[[folder/page-name]]`
 - Citation: `[[sources/SRC-YYYY-MM-DD-NNN]]`
+- Trajectory citation: `[[trajectories/TRJ-YYYY-MM-DD-NNN]]`
 
 ## Guardrails
 

package/docs/commands.md

@@ -13,10 +13,17 @@
 | `/wiki-status`   | Show wiki health                      |
 | `/wiki-digest`   | Daily/weekly summary                  |
 | `/wiki-retro`    | Save atomic insights from tasks        |
+| `/wiki-model`    | View/set the background-task model     |
+| `/wiki-trajectories` | Enable/disable agent working-memory (`on`/`off`, opt-in) |
+| `/wiki-record`   | Capture the completed task's trajectory (requires trajectories enabled) |
+| `/wiki-skills`   | Search distilled skills + past cases (requires trajectories enabled) |
 
 ## Extension Tools
 
-The extension registers 12 tools the LLM can call directly:
+The extension always registers 13 tools the LLM can call directly. The 3 agent-trajectory
+tools (`wiki_capture_trajectory`, `wiki_distill_skills`, `wiki_recall_skill`) are **opt-in,
+off by default** (issue #80) — registered only when `llm-wiki.trajectories` is enabled
+(`/wiki-trajectories on`).
 
 | Tool                  | Purpose                                     |
 | --------------------- | ------------------------------------------- |
@@ -32,6 +39,9 @@ The extension registers 12 tools the LLM can call directly:
 | `wiki_rebuild_meta`   | Force metadata rebuild                      |
 | `wiki_log_event`      | Record custom event                         |
 | `wiki_watch`          | Schedule auto-updates                       |
+| `wiki_capture_trajectory` | Capture the completed task's tool-call trajectory |
+| `wiki_distill_skills` | Batch undistilled trajectories for skill synthesis |
+| `wiki_recall_skill`   | Recall distilled skills + similar past cases |
 
 ## Workflows
 
@@ -52,3 +62,13 @@ The extension registers 12 tools the LLM can call directly:
 3. Synthesize answer with `[[wikilink]]` citations
 4. If novel: create analysis page via `wiki_ensure_page(type="analysis")`
 5. Extension auto-updates metadata
+
+### Task → Record → Distill (agent working-memory)
+
+_Opt-in: enable first with `/wiki-trajectories on`._
+
+1. Finish a non-trivial task (debug, refactor, integration)
+2. `wiki_capture_trajectory(title="...")` — auto-extracts the tool-call trajectory from the live session into `raw/trajectories/TRJ-*` with a self-contained summary (no skeleton to flesh)
+3. `wiki_distill_skills()` — get undistilled trajectories
+4. `wiki_ensure_page(type="skill", title="...")` — generalize into a reusable skill citing `[[trajectories/TRJ-...]]` (and optionally a `case` page)
+5. Next time, `wiki_recall_skill(query="...")` surfaces the skill/case before you start