geneseed 3.9.0 → 3.10.0

package/CHANGELOG.md

@@ -8,7 +8,38 @@ label. For the capability ↔ spec map, see [SHIPPED.md](SHIPPED.md).
 
 ## [Unreleased]
 
-_Nothing yet._
+## [3.10.0] — 2026-10-01
+
+**Upgrading:** re-emit after installing (`geneseed rebuild-all`). A Copilot install is no longer
+maintained: `geneseed uninstall` it before upgrading, or leave its files as they are — nothing
+re-emits them any more. Breaking for Copilot users only; every other host gains features.
+
+### Removed
+- **GitHub Copilot is no longer a host** — dropped for lack of support and of cross-tool
+  support. `--emit copilot` / `--emit copilot-global`, the `--host copilot` hook dialect, the
+  Copilot agent dialect, its MCP target, the setup and web-console options and
+  `adapters/copilot/` are gone. Geneseed now emits for OpenCode, Claude Code, Bob and
+  OpenClaude, plus the plain `AGENT.md` bundle (#166).
+
+### Fixed
+- **The session files reach the session.** The root file names `user-rules.md`, `PROFILE.md`,
+  the Memory and Notebook indexes, `wiki.jsonc` and `context.json` for session start, but on a
+  Claude-shaped install (Claude Code, Bob, OpenClaude) the context hook only looked at the repo
+  root — so none of them were ever injected, and the root told the model they sit "beside this
+  file", where they are not. The hook and the OpenCode context plugin now inject them from the
+  install's own folder as a `SESSION FILES` block, skipping any file still at its seed; the root
+  gives each host's real path; Claude-shaped installs now seed `context.json`; and the OpenCode
+  plugin finds `wiki.jsonc` on a project install (#167).
+- **Bob folder skills know where they live.** Bob hands a skill's text without its directory, so
+  a folder skill's `<this-skill-directory>/…` paths were unresolvable; the Bob emit now writes the
+  real path. `daydream` gained its missing frontmatter, `react-view-transitions` is named after
+  its folder, and `doctor` checks every folder skill's `name`/`description` (#164).
+
+### Changed
+- **A lighter lean root.** Restated prose is full-footprint only: the laws, Ethos, doctrine,
+  posture and mode intros, §3's dispatch envelope and council roster (they live in the
+  parallel-agents and council skills), §10 Scripts, and the folder-skill list on hosts that list
+  skills natively. The lean Claude root drops from ~33k to ~29k characters (#165).
 
 ## [3.9.0] — 2026-09-30
 

package/QUICKSTART.md

@@ -43,7 +43,7 @@ or a login item it did not create. Your clone keeps working meanwhile.
 
 ## Everything else
 
-Claude Code, Bob, Copilot, OpenClaude and plain-`AGENT.md` installs, per-repo mode, MCP servers, all
+Claude Code, Bob, OpenClaude and plain-`AGENT.md` installs, per-repo mode, MCP servers, all
 environment knobs, and troubleshooting live in the full **[Setup guide](SETUP.md)**.
 
 Once installed, point the agent at your repo's own docs with a tiny

package/README.md

@@ -14,7 +14,7 @@
 [![Agents](https://img.shields.io/badge/agents-18-orange)](src/agents/)
 [![Laws](https://img.shields.io/badge/laws-9-critical)](src/laws/universal.md)
 [![Plugins](https://img.shields.io/badge/plugins-7-teal)](adapters/opencode/plugins/)
-[![OpenCode · Claude Code · Bob · Copilot · OpenClaude · AGENT.md](https://img.shields.io/badge/works%20with-OpenCode%20·%20Claude%20Code%20·%20Bob%20·%20Copilot%20·%20OpenClaude%20·%20AGENT.md-1f6feb)](#-supported-harnesses)
+[![OpenCode · Claude Code · Bob · OpenClaude · AGENT.md](https://img.shields.io/badge/works%20with-OpenCode%20·%20Claude%20Code%20·%20Bob%20·%20OpenClaude%20·%20AGENT.md-1f6feb)](#-supported-harnesses)
 
 [**Why**](#-1--why-geneseed) · [**Setup**](#-2--setup) · [**Web & terminal**](#-3--web--terminal) · [**What you get**](#-4--what-you-get)
 
@@ -24,7 +24,7 @@
 
 <p align="center"><img src="docs/assets/demo.svg" alt="geneseed setup, then two laws catching a force-push and a credential at the tool boundary" width="760"></p>
 
-**Geneseed compiles your agent's rules instead of asking you to write them.** One source in `src/` renders into a ready-made harness for OpenCode, Claude Code, Bob, Copilot, OpenClaude, or any `AGENT.md` tool: a constitution of laws and doctrines, a roster of capability agents, native skills, a memory convention — and, wherever the host has a hook surface, gates that *enforce* the laws at the tool boundary instead of hoping the model remembers them.
+**Geneseed compiles your agent's rules instead of asking you to write them.** One source in `src/` renders into a ready-made harness for OpenCode, Claude Code, Bob, OpenClaude, or any `AGENT.md` tool: a constitution of laws and doctrines, a roster of capability agents, native skills, a memory convention — and, wherever the host has a hook surface, gates that *enforce* the laws at the tool boundary instead of hoping the model remembers them.
 
 ```bash
 npx geneseed setup
@@ -36,8 +36,8 @@ Want to read what it leaves behind before installing anything? [**geneseed-demo*
 
 A hand-written instructions file is prose the model may or may not honour, copied into every repo and drifting in each one. Geneseed is a build, and that changes four things:
 
-- **Laws are enforced, not suggested.** A force-push, a `reset --hard`, or a credential written into a tracked file is caught by a hook *before* the tool runs, in the host's own dialect — a prompt on Claude Code, Copilot and OpenClaude, a hard block on OpenCode, exit 2 on Bob. The gates fail closed, and every catch is one line in a ledger that `geneseed status` counts.
-- **One source, six targets.** Skills are byte-identical on every host; only the wiring differs. Every commit renders all 317 emit configurations, and a second emit into the same tree must change nothing.
+- **Laws are enforced, not suggested.** A force-push, a `reset --hard`, or a credential written into a tracked file is caught by a hook *before* the tool runs, in the host's own dialect — a prompt on Claude Code and OpenClaude, a hard block on OpenCode, exit 2 on Bob. The gates fail closed, and every catch is one line in a ledger that `geneseed status` counts.
+- **One source, five targets.** Skills are byte-identical on every host; only the wiring differs. Every commit renders all 261 emit configurations, and a second emit into the same tree must change nothing.
 - **Costs are measured, not guessed.** Zero runtime dependencies. The hook path loads in about 14 ms per tool call. The default *lean* footprint keeps the always-on context small and puts the full rationale one read away — the numbers are in [docs/token-footprint.md](docs/token-footprint.md).
 - **It follows you.** Install once, globally; every repo inherits it. One `git pull` or `npm install -g geneseed@latest` rebuilds every active install.
 
@@ -212,7 +212,7 @@ A second per-install dial, **footprint**, sets how much of the constitution `AGE
 
 Both footprints put the full text on disk beside `AGENT.md`: `laws/`, `ontology/` and `doctrines/` all ship in the bundle at **full** text whatever the footprint, and **all five pack files ship whether or not the pack was built in** — which is what lets a citation into an inactive pack resolve. So lean is a context/token optimization, **not** a rules cut.
 
-Lean is the default: the rationale is one read away and the context it frees is paid back on every turn. Switch to **full** when token cost is a non-issue or you run a smaller model, which leans harder on always-present rationale. Set it with `--footprint lean|full`, the Settings toggle, the per-harness dropdown in the Harnesses tab, or the setup wizard. It's remembered in a `.geneseed-footprint` marker and preserved across rebuilds, on every host (OpenCode, Claude Code, Bob, Copilot).
+Lean is the default: the rationale is one read away and the context it frees is paid back on every turn. Switch to **full** when token cost is a non-issue or you run a smaller model, which leans harder on always-present rationale. Set it with `--footprint lean|full`, the Settings toggle, the per-harness dropdown in the Harnesses tab, or the setup wizard. It's remembered in a `.geneseed-footprint` marker and preserved across rebuilds, on every host (OpenCode, Claude Code, Bob, OpenClaude).
 
 Either way the harness is otherwise identical — same files, Rules, capabilities, and guards; lean only relocates the *reasoning* to on-demand (and adds the standalone law, ontology and doctrine files to global/Claude/Bob installs). The one behavioural edge: with the rationale always in context, **full** applies a rule's nuance more reliably on subtle edge cases — or with a weaker model that may not reach for the pointer.
 
@@ -222,37 +222,37 @@ Want to check a build before it touches anything real? `geneseed validate --them
 
 ## 🔌 Supported harnesses
 
-One source, six emit targets. Geneseed builds into whichever host you point it
+One source, five emit targets. Geneseed builds into whichever host you point it
 at — each with a per-repo and a global (`-global`) variant — plus a portable
 `files` bundle any `AGENT.md`-aware tool can read. **OpenCode** runs its own
-engine (JS plugins, colour themes, LSP); **Claude Code**, **Bob**, **Copilot**
+engine (JS plugins, colour themes, LSP); **Claude Code**, **Bob**
 and **OpenClaude** share one Claude-shaped engine that diverges only by host dialect.
 
 The harness — its Rules, Agents, Skills, Memory convention, and preamble voice —
 is **identical on every host**. What differs is how much of it the host can
 *automate* for you (via plugins or hooks) versus carry as preamble discipline.
 
-| Capability | OpenCode | Claude Code | Bob | Copilot | OpenClaude |
-| --- | :---: | :---: | :---: | :---: | :---: |
-| **Instructions file** | `AGENT.md` + `opencode.json` | `CLAUDE.md` | `AGENTS.md` + `rules/geneseed.md` | `AGENTS.md` / `copilot-instructions.md` | `CLAUDE.md` (in `.openclaude/` per repo) |
-| **Agents** (capability specialists) | ✅ native | ✅ | ✅ | ✅ `.agent.md` | ✅ |
-| **Skills** (byte-identical) | ✅ | ✅ | ✅ | ✅ | ✅ |
-| **Memory & Notebook** | ✅ | ✅ | ✅ | ✅ | ✅ |
-| **Context injection** | ⚙️ plugin | 🪝 hook | 🪝 hook¹ | 🪝 hook³ | 🪝 hook⁴ |
-| **Memory write-back** (learn) | ⚙️ plugin | 🪝 hook | 🪝 hook¹ | 🪝 hook³ | 🪝 hook⁴ |
-| **Git-gate consent** (process 5) | ⚙️ plugin | 🪝 hook | 🪝 hook¹ (warn) | 🪝 hook³ | 🪝 hook⁴ |
-| **Rule-gate consent** (process 1) | ⚙️ plugin² | 🪝 hook | 🪝 hook¹ (warn) | 🪝 hook³ | 🪝 hook⁴ |
-| **Laws I / IV at the boundary** | ⚙️ plugin (block) | 🪝 hook (ask) | 🪝 hook¹ (exit 2) | 🪝 hook³ (ask) | 🪝 hook⁴ (ask) |
-| **Sovereign-repo excludes** | ⚙️ plugin | ✅ `claudeMdExcludes` | ✅ rules-shadow | ➖ none | ✅ `claudeMdExcludes` |
-| **MCP server wiring** | ✅ `mcp` | ✅ `mcpServers` | ✅ `mcp.json` | ✅ `mcp-config.json` | ✅ `mcpServers` |
-| **Colour themes** | ✅ full palette | ➖ | ➖ | ➖ | ➖ |
-| **LSP · workflow runner · primary-agent · `/`-commands** | ✅ | ➖ | ➖ | ➖ | ➖ |
-
-<sub>✅ native support · ⚙️ OpenCode plugin · 🪝 `settings.json` hook · 📄 carried by preamble prose only · ➖ no host mechanism (harness discipline still applies) · ¹ Bob's own hook contract: global hooks in `~/.bob/settings/settings.json`, Claude's event names but stdout ignored on `PreToolUse` — a refusal is **exit code 2**. So Laws I/IV exit 2, the consent rules are a stderr line, `SessionStart` context is plain stdout, `Stop` runs learn; `SubagentStop`/`PreCompact` are not Bob events and are not written. Unverified live (no Bob install on the authoring machine); the harness still holds via the rules preamble. · ² OpenCode's `tool.execute.before` can only allow or throw, with no "ask the user" tier, so the rule gate is a one-shot speed bump there rather than a prompt. · ³ Copilot's `~/.copilot/settings.json` hooks (`sessionStart`, `preToolUse`, `agentStop`, `preCompact`) wired by the **global** emit only; `preToolUse` has an ask tier, so every rule asks as on Claude Code. Unverified live (no Copilot CLI on the authoring machine). · ⁴ OpenClaude is a Claude Code fork: Claude's hook groups and verdicts verbatim, in `~/.openclaude/settings.json` (global) or `.openclaude/settings.local.json` (per repo). Unverified live (no OpenClaude install on the authoring machine).</sub>
+| Capability | OpenCode | Claude Code | Bob | OpenClaude |
+| --- | :---: | :---: | :---: | :---: |
+| **Instructions file** | `AGENT.md` + `opencode.json` | `CLAUDE.md` | `AGENTS.md` + `rules/geneseed.md` | `CLAUDE.md` (in `.openclaude/` per repo) |
+| **Agents** (capability specialists) | ✅ native | ✅ | ✅ | ✅ |
+| **Skills** (byte-identical) | ✅ | ✅ | ✅ | ✅ |
+| **Memory & Notebook** | ✅ | ✅ | ✅ | ✅ |
+| **Context injection** | ⚙️ plugin | 🪝 hook | 🪝 hook¹ | 🪝 hook³ |
+| **Memory write-back** (learn) | ⚙️ plugin | 🪝 hook | 🪝 hook¹ | 🪝 hook³ |
+| **Git-gate consent** (process 5) | ⚙️ plugin | 🪝 hook | 🪝 hook¹ (warn) | 🪝 hook³ |
+| **Rule-gate consent** (process 1) | ⚙️ plugin² | 🪝 hook | 🪝 hook¹ (warn) | 🪝 hook³ |
+| **Laws I / IV at the boundary** | ⚙️ plugin (block) | 🪝 hook (ask) | 🪝 hook¹ (exit 2) | 🪝 hook³ (ask) |
+| **Sovereign-repo excludes** | ⚙️ plugin | ✅ `claudeMdExcludes` | ✅ rules-shadow | ✅ `claudeMdExcludes` |
+| **MCP server wiring** | ✅ `mcp` | ✅ `mcpServers` | ✅ `mcp.json` | ✅ `mcpServers` |
+| **Colour themes** | ✅ full palette | ➖ | ➖ | ➖ |
+| **LSP · workflow runner · primary-agent · `/`-commands** | ✅ | ➖ | ➖ | ➖ |
+
+<sub>✅ native support · ⚙️ OpenCode plugin · 🪝 `settings.json` hook · 📄 carried by preamble prose only · ➖ no host mechanism (harness discipline still applies) · ¹ Bob's own hook contract: global hooks in `~/.bob/settings/settings.json`, Claude's event names but stdout ignored on `PreToolUse` — a refusal is **exit code 2**. So Laws I/IV exit 2, the consent rules are a stderr line, `SessionStart` context is plain stdout, `Stop` runs learn; `SubagentStop`/`PreCompact` are not Bob events and are not written. Unverified live (no Bob install on the authoring machine); the harness still holds via the rules preamble. · ² OpenCode's `tool.execute.before` can only allow or throw, with no "ask the user" tier, so the rule gate is a one-shot speed bump there rather than a prompt. · ³ OpenClaude is a Claude Code fork: Claude's hook groups and verdicts verbatim, in `~/.openclaude/settings.json` (global) or `.openclaude/settings.local.json` (per repo). Unverified live (no OpenClaude install on the authoring machine).</sub>
 
 **Reading the matrix.** Everything above the divider is at full parity — no host
 drops an Agent, Skill, or the memory convention. The asymmetry is entirely in
-*automation mechanism*: OpenCode's plugin surface and the Claude/Bob/Copilot/OpenClaude hook
+*automation mechanism*: OpenCode's plugin surface and the Claude/Bob/OpenClaude hook
 surfaces enforce a few Rules for you, each in the tier its host offers (a prompt on
 Claude Code, a hard block or a logged warning where the host has no prompt to
 give), and what no host can automate rides the preamble. The OpenCode-only extras (themes, LSP,
@@ -260,7 +260,7 @@ workflow runner, primary-agent) have no analogue on a Claude-shaped host.
 
 Per-host wiring in depth: **[OpenCode](adapters/opencode/README.md)** ·
 **[Claude Code](adapters/claude-code/README.md)** ·
-**[Bob](adapters/bob/README.md)** · **[Copilot](adapters/copilot/README.md)** ·
+**[Bob](adapters/bob/README.md)** ·
 **[OpenClaude](adapters/openclaude/README.md)**.
 Token cost per host: **[docs/token-footprint.md](docs/token-footprint.md)**.
 
@@ -294,7 +294,7 @@ Geneseed/
 ├── tests/                Node test suites, golden.mjs (every emit config) and mutate.mjs
 ├── docs/                 guides (web-ui, wiki, …) + docs/web/ (the console's Docs pages);
 │                         specs/, reviews/, superpowers/ are local working docs — git-ignored
-├── adapters/             per-host glue (opencode/, claude-code/, bob/, copilot/, openclaude/)
+├── adapters/             per-host glue (opencode/, claude-code/, bob/, openclaude/)
 └── .github/workflows/    ci.yml (doctor + tests) · publish.yml (npm, OIDC, manual only)
 ```
 
@@ -345,7 +345,6 @@ Details and precedence rules: [SETUP.md → Upgrade](SETUP.md#upgrade).
 | ⤷ [HOW-OPENCODE-LOADS.md](adapters/opencode/HOW-OPENCODE-LOADS.md) | Why a file shows up twice; plugin loading |
 | **[adapters/claude-code/](adapters/claude-code/README.md)** | The Claude Code hook adapter |
 | **[adapters/bob/](adapters/bob/README.md)** | The IBM Bob adapter — Claude-shaped, rules-file preamble |
-| **[adapters/copilot/](adapters/copilot/README.md)** | The GitHub Copilot adapter — global hooks (`sessionStart`, `preToolUse`, `agentStop`, `preCompact`) |
 | **[adapters/openclaude/](adapters/openclaude/README.md)** | The OpenClaude adapter — the Claude engine under `~/.openclaude` / `.openclaude/` |
 | **[src/memory/README.md](src/memory/README.md)** | The memory convention |
 | **[src/notebook/README.md](src/notebook/README.md)** | The agent's own freeform-space convention |

package/SETUP.md

@@ -22,7 +22,7 @@ that matches your tool, then configure and verify. For the conceptual overview s
 
 - **Node ≥ 22.3** — the only hard requirement. Geneseed ships as an npm package with
   **zero dependencies**; `npx geneseed` needs nothing else.
-- **Your agent tool** — OpenCode (recommended), Claude Code, Bob, Copilot, OpenClaude, or anything
+- **Your agent tool** — OpenCode (recommended), Claude Code, Bob, OpenClaude, or anything
   that reads a root instructions file.
 - *Only for a git checkout:* **git**. Nothing else — a checkout runs the same Node
   entry points the npm package installs.
@@ -93,8 +93,7 @@ hand? Pick a path below.
 | [A — OpenCode, global](#path-a--opencode-global-recommended) | **Recommended.** One install, every repo inherits it, nothing committed into projects. |
 | [B — OpenCode, per-repo](#path-b--opencode-per-repo) | You want a committed `.opencode/` layer in one repository. |
 | [C — Claude Code](#path-c--claude-code) | You drive Claude Code and want the lifecycle hooks. |
-| [C′ — GitHub Copilot](#path-c--github-copilot) | You drive the Copilot CLI, coding agent, or VS Code agent mode. |
-| [C″ — OpenClaude](#path-c--openclaude) | You drive OpenClaude, the any-model Claude Code fork. |
+| [C′ — OpenClaude](#path-c--openclaude) | You drive OpenClaude, the any-model Claude Code fork. |
 | [D — Any `AGENT.md` tool](#path-d--any-agentmd-tool) | Cursor, Aider, or any tool that reads a root instructions file. |
 | [E — No runtime on the target at all](#path-e--no-runtime-on-the-target-at-all) | The machine that *uses* the harness cannot run Node. |
 
@@ -219,45 +218,11 @@ repo's own `.claude/settings.local.json`, and OpenClaude's in `.openclaude/setti
 Bob: the workspace rules shadow stub;
 OpenCode: the plugins stand down). Everything is reversed by `exclude remove`.
 
-Limitations: GitHub Copilot has no per-repo suppression mechanism, so the global
-`copilot-instructions.md` still loads there; and globally installed skills/
-subagents remain listed by the host (no native per-repo disable exists). A
-global install emitted later starts with an empty list — re-run
+Limitation: globally installed skills/subagents remain listed by the host (no
+native per-repo disable exists). A global install emitted later starts with an empty list — re-run
 `harness exclude add` (`exclude list` flags installs that diverge).
 
-### Path C′ — GitHub Copilot
-
-```
-geneseed-build --emit copilot-global           # personal: render into ~/.copilot
-geneseed-build --emit copilot --out . --root .  # per-repo: AGENTS.md + .github/, committed
-```
-
-Copilot is Claude-shaped where it counts: skills are the same `SKILL.md` folders
-(Copilot's Agent Skills — `.github/skills/` in a repo, `~/.copilot/skills/`
-personally), agents render in Copilot's own custom-agent dialect
-(`agents/<name>.agent.md`, a `tools:` allowlist instead of Claude's denylist), and
-the preamble rides a file Copilot auto-loads: the repo-root `AGENTS.md` (CLI, coding
-agent, and VS Code agent mode) or the personal `~/.copilot/copilot-instructions.md`
-(CLI). Two differences from the Claude paths:
-
-- **Hooks at global scope only.** The global emit adds entries to four events in
-  `~/.copilot/settings.json`: `sessionStart` (eager project-context injection),
-  `preToolUse` (both gates behind one command, since entries carry no matcher —
-  every rule **asks**, as on Claude Code), and `agentStop` + `preCompact` (`learn`,
-  from the payload's `transcriptPath`). The per-repo emit wires nothing: a
-  machine-absolute command in a shared `.github/` would fail on every teammate's
-  machine.
-- **The per-repo layer lives in the shared `.github/`** (Copilot's repo config
-  surface). Safe by construction: the ownership manifest + claim-on-create never
-  touch files Geneseed didn't write — your workflows and same-named agents/skills
-  survive every emit and uninstall.
-
-MCP servers go in `~/.copilot/mcp-config.json` (the Settings/MCP screens know the
-shape). `$COPILOT_CONFIG_DIR` relocates the personal dir, mirroring
-`$BOB_CONFIG_DIR`. Note both carriers stack if you install globally *and* per-repo —
-the global emit warns when that's about to happen.
-
-### Path C″ — OpenClaude
+### Path C′ — OpenClaude
 
 ```
 geneseed-build --emit openclaude-global           # render into ~/.openclaude
@@ -453,7 +418,7 @@ you run a smaller model.
 **Same harness, either way.** Footprint changes neither what the harness *is* nor what it
 can *do*: lean and full emit identical files (same agents, skills, plugins, commands, memory,
 notebook, hooks) and every Rule is present and binding. The only structural difference is
-that a lean install on a global / Claude / Bob / Copilot / OpenClaude target also ships the standalone
+that a lean install on a global / Claude / Bob / OpenClaude target also ships the standalone
 `laws/universal.md` and `ontology/` (project bundles already carry them, and `doctrines/`
 ships at both footprints on every target); the only behavioural difference is that the
 reasoning loads on demand instead of every turn. Lean is the default; full, with the rationale
@@ -463,7 +428,7 @@ with a weaker model, and is one flag away.
 Set it with `--footprint lean|full` (alongside any `--emit`), the **Footprint** toggle in
 the web Settings, the per-harness dropdown in the Harnesses tab, or the setup wizard. It is
 remembered in a `.geneseed-footprint` marker and preserved across every rebuild, on every
-host (OpenCode, Claude Code, Bob, Copilot, OpenClaude).
+host (OpenCode, Claude Code, Bob, OpenClaude).
 
 ### Dry-run a build (`validate`)
 

package/SHIPPED.md

@@ -20,7 +20,7 @@ historical references, not files distributed with the repo.
 > **On version numbers.** [`harness.config.json`](harness.config.json)'s `version` is a
 > human-readable label only. The canonical identity of an installed harness is the
 > **source fingerprint** stamped in `.geneseed-version` (see `geneseed version`), not the
-> string. Current label: **3.9.0**.
+> string. Current label: **3.10.0**.
 
 ## Capabilities
 
@@ -37,7 +37,7 @@ Every row below is present in the tree today; "Spec" names the design record.
 | **Terminal rendering** | There is no full-screen control panel: `tui` and `menu` are verbs that say so and print the command list. The display-width and glyph layer ships on, behind `status` and the install animation | `2026-06-07-tui-refactor-polish` · `2026-06-08-tui-modern-refresh` · `2026-06-11-windows-tui` |
 | **Web console** | Local browser UI (Dashboard; Codex: Constitution, Rules, Profile, Skills, Agents, Library, Docs; Care: Activity, Changes, Doctor; Setup: Harness, Settings), auto-build, in-app docs menu | `2026-06-12-web-ui-v2` · `2026-06-12-web-auto-build` · `2026-06-13-docs-menu` |
 | **OpenCode adapter** | Native agents/skills mapping + plugins (context, learn, guard, workflow, notify, ponytail, activity); global-emit link integrity; JSONC config target | `2026-06-07-opencode-adapter-features` · `2026-06-07-global-emit-link-integrity` · `2026-06-12-opencode-jsonc-target` · `2026-06-16-ponytail-minimal-code-mode` |
-| **Host hooks, per dialect** | Every host gets the hook surface its runtime offers, spoken in that runtime's own verdict dialect via `--host`: Claude Code asks (`permissionDecision`), OpenClaude asks exactly as Claude does (a Claude Code fork, under `~/.openclaude` / `.openclaude/`), Copilot asks as Claude does (`sessionStart`, `preToolUse`, `agentStop`, `preCompact` in `~/.copilot/settings.json`, global scope only), Bob refuses by exit code 2 (five Claude-named events, nested `~/.bob/settings/settings.json`, no subagent event), OpenCode's guard throws. Laws I/IV are the hard tier everywhere; the consent rules ask where a host can ask and warn where it cannot. `tool-gate` fuses the two gates for hosts that run one command per event; `PreCompact` runs `learn` before Claude compacts; every block or ask lands in `notebook/gates.jsonl` on every host | [copilot](adapters/copilot/README.md) · [bob](adapters/bob/README.md) · [claude-code](adapters/claude-code/README.md) · [candidate hosts](docs/candidate-hosts.md) |
+| **Host hooks, per dialect** | Every host gets the hook surface its runtime offers, spoken in that runtime's own verdict dialect via `--host`: Claude Code asks (`permissionDecision`), OpenClaude asks exactly as Claude does (a Claude Code fork, under `~/.openclaude` / `.openclaude/`), Bob refuses by exit code 2 (five Claude-named events, nested `~/.bob/settings/settings.json`, no subagent event), OpenCode's guard throws. Laws I/IV are the hard tier everywhere; the consent rules ask where a host can ask and warn where it cannot. `tool-gate` fuses the two gates for hosts that run one command per event; `PreCompact` runs `learn` before Claude compacts; every block or ask lands in `notebook/gates.jsonl` on every host | [bob](adapters/bob/README.md) · [claude-code](adapters/claude-code/README.md) · [candidate hosts](docs/candidate-hosts.md) |
 | **Workflow primitive** | Saved, code-driven subagent orchestration (`workflow` tool + runtime); shipped scripts include `dispatch` — decompose a goal, route each subtask to its owning agent, converge | `2026-06-09-opencode-workflow-primitive` |
 | **Per-agent memory** | Each capability agent keeps durable lessons in `memory/agents/<name>.md`: read at dispatch, written back mechanically by `learn` on session/subagent end (OpenCode plugin + Claude `SubagentStop`; Bob has no subagent event); `learn --consolidate` rebuilds the index | `2026-07-04-orchestration-memory-upgrade` |
 | **Wiki integration** | Machine-wide knowledge base (`wiki.jsonc`): eager/lazy loading, conventions, protected folders enforced at the tool boundary | `2026-06-11-wiki-knowledge-base` |

package/adapters/bob/README.md

@@ -23,7 +23,8 @@ limitation to be aware of on a shared repo.
   - `agents/<name>.md` — the **Claude subagent dialect** verbatim (`name` +
     `description` + a `disallowedTools:` denylist for read-only agents).
   - `skills/<name>/SKILL.md` — **byte-identical** to every other host; skills
-    are model-invoked via the `skill` tool.
+    are model-invoked via the `skill` tool. The one exception is the folder skills'
+    own directory — see *Bob-ism 3*.
   - `rules/geneseed.md` — a **slim shadow stub** (see *Bob-ism 1* below).
   - `settings.json` — Geneseed's lifecycle hooks (gitignored by the emit's
     `.bob/.gitignore`, see the caveat below).
@@ -64,6 +65,21 @@ never gets one; the `rules/geneseed.md` shadow above covers the same
 project-bypasses-global need. An older Bob install that wrote a stale exclude is
 **self-healed** (the entry is removed) on the next emit.
 
+## Bob-ism 3 — a folder skill is told where it lives
+
+A folder skill (`explain-changes`, `token-report`, `daydream`, …) ships scripts or
+prompts beside its `SKILL.md` and names them `<this-skill-directory>/…`. Claude Code
+answers that placeholder itself — it gives the model each skill's base directory on load.
+Bob gives the skill's text only, so the model was left with a path it could not resolve
+and fell back to searching the filesystem. The Bob emit therefore writes the directory in:
+
+- per repo, relative to the workspace root — `node .bob/skills/token-report/scripts/…` —
+  which is where Bob runs commands, and stays valid in a teammate's checkout;
+- globally, the absolute path under `~/.bob/skills/`.
+
+`doctor` also requires every folder skill's `SKILL.md` to open with a frontmatter whose
+`name` is the folder's and which has a `description` — what any host registers a skill from.
+
 ## Hooks — Bob's own protocol
 
 Bob documents lifecycle hooks (its hooks page and the 2.0.2 changelog; see

package/adapters/openclaude/README.md

@@ -58,7 +58,7 @@ which root file OpenClaude already loads (`AGENTS.md` when present, else
 ## Not supported
 
 - **Workflows.** OpenClaude ships its `WorkflowTool` disabled, so the `workflow`
-  skill falls back to parallel agents, as on Bob and Copilot.
+  skill falls back to parallel agents, as on Bob.
 - **Live verification.** The emit follows OpenClaude's source (`src/utils/envUtils.ts`,
   `src/utils/claudemd.ts`, `src/utils/config.ts`, `src/types/hooks.ts`) as of
   2026-09. It has not been run against an installed OpenClaude yet.

package/adapters/opencode/plugins/geneseed-context.js

@@ -56,6 +56,7 @@ import { promises as fs } from "node:fs"
 import * as path from "node:path"
 import { fileURLToPath } from "node:url"
 import { homedir } from "node:os"
+import { createHash } from "node:crypto"
 
 const MARKER = "<!-- geneseed-context:v2 -->"
 const PLUGIN_DIR = path.dirname(fileURLToPath(import.meta.url))
@@ -156,6 +157,49 @@ const EXCLUDE_DIRS = new Set([
 async function isFile(p) { try { return (await fs.stat(p)).isFile() } catch { return false } }
 async function isDir(p) { try { return (await fs.stat(p)).isDirectory() } catch { return false } }
 
+// ---- session files (the harness's own) --------------------------------------
+// The files AGENT.md names for session start, relative to the HARNESS dir — the dir that
+// holds the installed AGENT.md: the repo root for a project install, the config dir for a
+// global one. Same list and seed hashes as SESSION_FILES/SEED_SHA256 in js/hosts/hooks.mjs;
+// tests/unit/claude.test.mjs gates both copies against SESSION_SEEDS in js/build/stubs.mjs.
+// context.json is not here: it is a manifest, honoured by resolveSource. wiki.jsonc is not
+// here either: the MACHINE WIKI block below already renders it in full.
+const SESSION_FILES = ["user-rules.md", "PROFILE.md", "memory/MEMORY.md", "anamnesis/MEMORY.md",
+  "notebook/NOTEBOOK.md"]
+const SEED_SHA256 = new Set([
+  "5c2e92fb1acde041e02d9ebf158460d8ff7a07cbc2b447859631604610130721", // user-rules.md
+  "29e012c3c4349ea62a9082cbd04bb25599a0ec280d62c33f6e10742809471816", // PROFILE.md
+  "cccc917c34e6b990e821620bdb4739090151885ab71b6e85046f30a3b71fa5e8", // wiki.jsonc
+  "99c786049c260f6baf7aec68a5c0f59907861afe9b867da009440613bdddb907", // MEMORY.md
+  "9acb5c9d9cb5dac572104480f05b48cc144e676869bd60e86d8a15b1f6308184", // NOTEBOOK.md
+])
+
+// $GENESEED_HARNESS when set, else the first of <plugin dir>/.. (global: the config dir) and
+// <plugin dir>/../.. (project: .opencode/plugins -> the repo root) that holds AGENT.md.
+async function harnessDir() {
+  if (process.env.GENESEED_HARNESS && (await isDir(process.env.GENESEED_HARNESS))) {
+    return process.env.GENESEED_HARNESS
+  }
+  for (const d of [path.resolve(PLUGIN_DIR, ".."), path.resolve(PLUGIN_DIR, "..", "..")]) {
+    if (await isFile(path.join(d, "AGENT.md"))) return d
+  }
+  return null
+}
+
+// Session files present under the harness dir and changed from their seed: [{ rel, abs, text }].
+async function sessionFiles(dir) {
+  const found = []
+  if (!dir) return found
+  for (const rel of SESSION_FILES) {
+    const abs = path.join(dir, rel)
+    let text
+    try { text = (await fs.readFile(abs, "utf8")).replace(/\r\n/g, "\n") } catch { continue }
+    if (SEED_SHA256.has(createHash("sha256").update(text).digest("hex"))) continue
+    found.push({ rel, abs, text: text.replace(/\n+$/, "") })
+  }
+  return found
+}
+
 async function readJson(p) {
   try { return JSON.parse(await fs.readFile(p, "utf8")) } catch { return null }
 }
@@ -474,11 +518,13 @@ async function fromManifest(manifestFile, root) {
   return { eager: eager.sort(byRel), lazy: lazy.sort(byRel) }
 }
 
-async function resolveSource(root) {
+async function resolveSource(root, harness = null) {
   const explicit = process.env.GENESEED_CONTEXT
   if (explicit && (await isFile(explicit))) return { mode: "manifest", file: explicit }
-  for (const rel of [path.join(".harness", "context.json"), "context.json"]) {
-    const p = path.join(root, rel)
+  // The harness dir last: a repo's own manifest outranks the one seeded beside the install.
+  const cands = [path.join(root, ".harness", "context.json"), path.join(root, "context.json")]
+  if (harness) cands.push(path.join(harness, "context.json"))
+  for (const p of cands) {
     if (await isFile(p)) return { mode: "manifest", file: p }
   }
   return { mode: "discover" }
@@ -494,8 +540,12 @@ async function resolveSource(root) {
 async function resolveWikiFile() {
   const explicit = process.env.GENESEED_WIKI
   if (explicit && (await isFile(explicit))) return explicit
+  // The harness dir (which honours $GENESEED_HARNESS) first: on a PROJECT install the plugin
+  // sits in .opencode/plugins, so <plugin dir>/.. is .opencode, not the dir the seed landed in.
+  // <plugin dir>/.. stays as the fallback for an install without an AGENT.md.
   const bases = []
-  if (process.env.GENESEED_HARNESS) bases.push(process.env.GENESEED_HARNESS)
+  const harness = await harnessDir()
+  if (harness) bases.push(harness)
   bases.push(path.resolve(PLUGIN_DIR, ".."))
   for (const base of bases) {
     // wiki.json is the legacy name from earlier seeds — still honoured.
@@ -652,9 +702,27 @@ async function renderSet(out, { eager, lazy }, state) {
   state.lazy += lazyLines.length
 }
 
-async function buildBlock(sets, wikis = [], commands = []) {
+async function buildBlock(sets, wikis = [], commands = [], session = []) {
   const state = { spent: 0, injected: 0, lazy: 0, headingsRead: 0 }
 
+  // A session file discovery also found (user-rules.md at the root of a project install,
+  // whose harness dir IS the repo root) is injected once, in the session block.
+  const seen = new Set(session.map((s) => path.resolve(s.abs)))
+  const fresh = (x) => !seen.has(path.resolve(x.abs))
+  sets = { eager: sets.eager.filter(fresh), lazy: sets.lazy.filter(fresh) }
+  const ses = []
+  for (const s of session) {
+    const size = Buffer.byteLength(s.text, "utf8")
+    ses.push(`----- ${s.rel} -----`)
+    if (size > EAGER_FILE_KB * 1024) {
+      ses.push(`[demoted: ${s.rel} exceeded ${EAGER_FILE_KB} KB — read ${s.abs} on demand]`, "")
+      continue
+    }
+    ses.push(s.text, "")
+    state.spent += size
+    state.injected++
+  }
+
   const proj = []
   await renderSet(proj, sets, state)
 
@@ -675,8 +743,9 @@ async function buildBlock(sets, wikis = [], commands = []) {
     if (w.truncated) wik.push(`  [+${w.truncated} more notes in this wiki — explore its folders on demand]`, "")
   }
 
-  if (!proj.length && !wik.length) return null
+  if (!ses.length && !proj.length && !wik.length) return null
   const out = [MARKER]
+  if (ses.length) out.push("=== SESSION FILES — your harness files, injected at session start (untouched seeds skipped) ===", "", ...ses)
   if (proj.length) out.push("=== PROJECT CONTEXT — binding for this repo ===", "", ...proj)
   if (wik.length) out.push("=== MACHINE WIKI — the user's knowledge base, binding per AGENT.md §8 ===", "", ...wik)
   return { text: out.join("\n"), injected: state.injected, lazy: state.lazy, kb: kb(state.spent) }
@@ -686,13 +755,14 @@ async function buildBlock(sets, wikis = [], commands = []) {
 // and render the injection block — shared by the session.created injection and the
 // compaction hook (the transform path reuses it via cachedBlockText).
 async function resolveBlock(root) {
-  const src = await resolveSource(root)
+  const harness = await harnessDir()
+  const src = await resolveSource(root, harness)
   const sets = src.mode === "manifest"
     ? await fromManifest(src.file, root)
     : await discover(root)
   const wikis = await wikiSets()
   const commands = await discoverCommands(root)
-  return { block: await buildBlock(sets, wikis, commands), src }
+  return { block: await buildBlock(sets, wikis, commands, await sessionFiles(harness)), src }
 }
 
 // Per-process cache of the rendered block text for the transform path (root is fixed

package/bin/geneseed-hook.mjs

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 /**
  * The hook entry point — a Node twin of `rituals/harness.py` for the verbs the emitted
- * `settings.json` actually invokes (four from the reference, plus `tool-gate` for Copilot).
+ * `settings.json` actually invokes (four from the reference, plus `tool-gate` for Bob).
  *
  * WHY IT IS A SECOND BINARY AND NOT A SUBCOMMAND OF `bin/build-driver.mjs`. Three reasons,
  * and each alone would be enough:

package/docs/candidate-hosts.md

@@ -38,8 +38,7 @@ current host gets. Skills are the largest surface by file count and they cost no
   Claude's, and `permissionDecision: "ask"` exists. The Claude engine with `host: "codex"`
   covers instructions (`AGENTS.md`), skills (`.agents/skills`) and the hook verbs
   unchanged; the only new code is the carrier writer (a `hooks.json` merge or a TOML
-  table) and the config-dir resolver (`~/.codex`, `$CODEX_HOME`). Roughly the Copilot
-  port minus the dialect work.
+  table) and the config-dir resolver (`~/.codex`, `$CODEX_HOME`).
 
 - **Gemini CLI — Claude's groups, Gemini's names.** The hook groups are Claude's
   `event → [{matcher, hooks}]` and the payload fields are Claude's, but the events are
@@ -54,7 +53,7 @@ current host gets. Skills are the largest surface by file count and they cost no
 - **Cursor — a third verdict dialect.** `AGENTS.md` and `.agents/skills` are free; hooks
   are `hooks.json` with a `{command}` list per event and a `permission: allow|deny`
   verdict, `additional_context` in snake_case for `sessionStart`. Same block-or-warn stance
-  as Copilot, one more output spelling in `js/hosts/hooks.mjs`, one more carrier writer.
+  as Bob, one more output spelling in `js/hosts/hooks.mjs`, one more carrier writer.
   Cursor also reads `.claude/skills/`, so a Claude project install already gives a Cursor
   user the skills.
 
@@ -64,8 +63,16 @@ If a fifth host is wanted, Codex first: it reuses the most and it is the only ca
 where the gates can *ask* rather than block. Before any of them, a `--emit agents` for
 the shared `.agents/skills/` folder would give Codex, Cursor and Gemini users the whole
 skill catalogue for the price of one directory name, with no host-specific code at all.
-A Gemini host would be the fourth verdict spelling in `js/hosts/hooks.mjs` (after Claude's
-ask, Copilot's block, Bob's exit code 2), plus a resolver and a carrier.
+A Gemini host would be the third verdict spelling in `js/hosts/hooks.mjs` (after Claude's
+ask and Bob's exit code 2), plus a resolver and a carrier.
+
+## Evaluated, dropped
+
+- **GitHub Copilot** — shipped as a host through 3.9.0, removed in 3.10.0. Dropped for lack
+  of support: its hooks were never verified live, its `postToolUse` output does not reach
+  the model, and it shared no cross-tool surface with the other hosts (camelCase payloads,
+  matcherless hook entries, its own agent dialect and carrier), so every feature needed a
+  Copilot-only decision.
 
 ## Not verified
 

package/docs/token-footprint.md

@@ -15,33 +15,33 @@ heavier on the root file.
 
 The context every host injects at session start, before you type anything:
 
-| Component | Claude Code | OpenCode | Bob | Copilot |
-|---|---|---|---|---|
-| Root instruction file | ~7.3k (`CLAUDE.md`) | ~7.3k (`AGENT.md`) | ~7.7k (`AGENTS.md`, §3 table inline) | ~9.0k (`AGENTS.md`, §3+§4 tables inline) |
-| Skill metadata (name + description) | ~4.1k | ~4.1k | ~4.1k (native) | — (§4 table; native catalogue unverified) |
-| Agent metadata | ~0.5k | ~0.5k | — (no agents directory) | — (§3 table) |
-| Eager memory/notebook injection | ~1.2k (SessionStart hook) | ~1.2k (context plugin) | ~1.2k (SessionStart hook) | ~1.2k (sessionStart hook) |
-| **Total** | **~13k** | **~13k** | **~13k** | **~10k** |
+| Component | Claude Code | OpenCode | Bob |
+|---|---|---|---|
+| Root instruction file | ~7.2k (`CLAUDE.md`) | ~7.2k (`AGENT.md`) | ~7.5k (`AGENTS.md`, §3 table inline) |
+| Skill metadata (name + description) | ~4.1k | ~4.1k | ~4.1k (native) |
+| Agent metadata | ~0.5k | ~0.5k | — (no agents directory) |
+| Session files (rules, profile, Memory/Notebook indexes, wiki) | 0 fresh, grows with use (SessionStart hook) | 0 fresh, grows with use (context plugin) | 0 fresh, grows with use (SessionStart hook) |
+| **Total** | **~12k** | **~12k** | **~12k** |
 
-The emits are at parity by design: ~10–13k tokens, about 6% of a 200k
+The emits are at parity by design: ~12k tokens, about 6% of a 200k
 window. Bob catalogues skills natively but has no agents directory, so its
 root keeps the §3 Agents table and drops the §4 Skills table — the catalogue
 flag is per kind (`hostCatalogsNatively`), not one boolean. Before the 2026-09 footprint pass every hooked host paid roughly double
 this: the context hook re-injected the root file the host had already loaded
 natively, and skill descriptions ran to 900 characters each — see the
-"Where the tokens go" list below for what changed. Bob and Copilot carry the §3/§4 catalogue tables inline because their
-hosts expose no native skill/agent inventory; Claude Code and OpenCode ship a
-pointer to the host's own inventory instead, ~1.9k tokens lighter. Copilot's
-eager injection rides its `sessionStart` hook, which the **global** emit wires;
-a project-scope Copilot install has no hooks, so there the memory/notebook
-indexes load when the agent reads them, not eagerly. The
+"Where the tokens go" list below for what changed. Bob carries the §3/§4 catalogue tables inline because its host exposes
+no native skill/agent inventory; Claude Code and OpenCode ship a
+pointer to the host's own inventory instead, ~1.9k tokens lighter. A session file still
+byte-identical to its seed is skipped — it says nothing — so a fresh install injects none
+of them; until 2026-10 the hook read them from the repo root, where a Claude-shaped install
+never puts them, and none reached the session at all. The
 eager-injection path is budget-capped identically everywhere — 16 KB per file
 (cut at a line break, with a marker saying so) and 48 KB per session (files past
 the budget are listed lazy with the reason) — so a 40k-character README no
 longer leaves whole every session, and growing Memory degrades every host the
 same way instead of one silently falling behind. The hook also **never injects
 the root file the host loads by itself** (`CLAUDE.md` on Claude Code,
-`AGENTS.md` on Bob and Copilot, `AGENT.md`/`AGENTS.md`/`CLAUDE.md` on OpenCode):
+`AGENTS.md` on Bob, `AGENT.md`/`AGENTS.md`/`CLAUDE.md` on OpenCode):
 until 2026-09 it did, and the whole harness was paid twice per session.
 
 ## Where the tokens go
@@ -72,10 +72,17 @@ until 2026-09 it did, and the whole harness was paid twice per session.
   capped at 320 characters (`skillDescription` in `js/hosts/native.mjs`); at
   the earlier 900-character cap the catalogue weighed ~5.6k tokens. The later
   trigger sentences still live in the body, read as soon as the skill is chosen.
-- **§5–§10 carry authored lean halves too** (Memory, Notebook, Workspace, Wiki,
-  Context, Scripts): each keeps its obligations — read `MEMORY.md`, read
+- **§5–§9 carry authored lean halves too** (Memory, Notebook, Workspace, Wiki,
+  Context): each keeps its obligations — read `MEMORY.md`, read
   `context.json` before the first reply, the git-ignored folders — and points at
-  the on-disk README or config file for the rest.
+  the on-disk README or config file for the rest. §10 (Scripts) is full-only: it
+  describes the CLI to a reader, not an obligation to the agent.
+- **Restated prose is full-only** (2026-10): the laws and Ethos intros, the §2
+  pack intro, the posture/mode intros, §3's dispatch envelope and council roster
+  (they live in the parallel-agents and council Skills; the lean root keeps
+  "dispatched agents never commit or push"), and the folder-Skill list, which
+  now sits in the §4 catalogue half and so drops wherever the host catalogues
+  Skills natively.
 - **Skill bodies (~53k) and agent bodies (~10k) are lazy** on every host —
   loaded only when invoked. A typical skill costs ≤3k per invocation; the
   heaviest (`react-view-transitions`, ~17.5k with its reference files) loads
@@ -97,9 +104,7 @@ token counter also includes, none of which Geneseed controls:
    definitions. On OpenCode this is typically 5–10k tokens before any
    harness content loads.
 2. **Your repo's docs, injected eagerly** — the context delivery
-   (plugin on OpenCode, SessionStart hook on Claude Code / Bob, `sessionStart`
-   hook on a global Copilot install; a project-scope Copilot install has no
-   hooks, so it reads docs on demand instead) discovers and
+   (plugin on OpenCode, SessionStart hook on Claude Code / Bob / OpenClaude) discovers and
    injects `README.md`, `CONTRIBUTING.md`, and files under `docs/`, up to the
    48 KB budget (≈12k tokens). A doc-heavy repo fills it.
 3. **Wiki eager entries** plus the lazy listing of the rest, if a wiki is

package/docs/web/claude-hooks.md

@@ -11,7 +11,7 @@ Claude Code has no `instructions` array and no JS plugin dir — the harness rea
 - **Context injection** — `SessionStart` (startup, clear, resume) runs `geneseed-hook context`: auto-discovers the repo's docs by convention and injects them before your first turn, plus your machine wiki. It honours the same `GENESEED_CONTEXT` manifest as the OpenCode context plugin.
 - **Git gate** — `PreToolUse` on Bash runs `geneseed-hook git-gate`: a `git commit` or `git push` anywhere in the command line — a compound one-liner and a `-C <path>` form included — asks first, at the tool boundary and before the shell runs it (**Doctrine process 5**). It is the one hook that rides a doctrine pack: an install built without **process** never gets this group, because a boundary that keeps asking for a rule the install did not adopt is a gate arguing with its own harness.
 - **Rule gate** — `PreToolUse` on the write tools runs `geneseed-hook rule-gate`: a write to `user-rules.md` or to a memory file asks first, because whether something you want kept is a standing rule or a durable fact is *your* call, settled through the [rule skill](#/docs/skills) (**Doctrine process 1**). Ordinary edits are untouched.
-- **Tool gate** — `geneseed-hook tool-gate` is the git gate and the rule gate fused behind one command, for hosts whose hook config has no per-tool matcher (Copilot, Bob). Claude Code matches per tool, so its `settings.json` wires the two gates above instead.
+- **Tool gate** — `geneseed-hook tool-gate` is the git gate and the rule gate fused behind one command, for hosts whose hook config has no per-tool matcher (Bob). Claude Code matches per tool, so its `settings.json` wires the two gates above instead.
 - **Learn** — `Stop` and `SubagentStop` run `geneseed-hook learn`: distils durable memories into the install's `memory/` store at session end; a subagent's stop routes to the per-agent lesson path (`memory/agents/<name>.md`).
 
 ### Where they live