@lemoncode/lemony 0.1.0 → 0.1.1-alpha.1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lemoncode/lemony",
-  "version": "0.1.0",
+  "version": "0.1.1-alpha.1",
   "description": "Lemony — a Harness for AI Coding. Vendor package: installer, agent role catalog, generic skill catalog, hooks, and templates for a Spec-Driven Development workflow.",
   "type": "module",
   "private": false,
@@ -16,8 +16,15 @@
   "files": [
     "dist",
     "catalog",
+    "!catalog/agents/README.md",
+    "!catalog/hooks/README.md",
+    "!catalog/playbook-format.md",
+    "!catalog/schemas/README.md",
+    "!catalog/skills/README.md",
+    "!catalog/templates/README.md",
     "README.md",
-    "PRIVACY.md"
+    "PRIVACY.md",
+    "NOTICE"
   ],
   "engines": {
     "node": ">=24",
@@ -49,6 +56,8 @@
     "format:check": "prettier --check .",
     "schema:generate": "tsx scripts/generate-config-schema.ts",
     "schema:check": "tsx scripts/generate-config-schema.ts --check",
+    "notice:generate": "tsx scripts/generate-notice.ts",
+    "notice:check": "tsx scripts/generate-notice.ts --check",
     "telemetry:aggregate": "tsx scripts/aggregate-telemetry.ts",
     "changeset": "changeset",
     "changeset:version": "changeset version && tsx scripts/sync-catalog-version.ts",
@@ -59,8 +68,8 @@
     "@stryker-mutator/core": "9.6.1",
     "@stryker-mutator/vitest-runner": "9.6.1",
     "@types/node": "24.13.2",
-    "oxlint": "1.70.0",
-    "prettier": "3.8.4",
+    "oxlint": "1.71.0",
+    "prettier": "3.9.1",
     "tsdown": "0.22.3",
     "tsx": "4.22.4",
     "typescript": "6.0.3",

package/catalog/agents/README.md

@@ -1,29 +0,0 @@
-# agents/ — role catalog templates
-
-> **Status: 1 hat + 5 sub-agents, all skill-gated (P4 slice 2).** `orchestrator` (hat)
-> weaves its skills into its prose. The sub-agents `spec-author`, `implementer`,
-> `reviewer`, and the on-demand `architect` each carry a single `{{SKILLS}}` marker the
-> installer fills with the skills they invoke, gated by `applies-when` capabilities
-> (decision #31; the profile tier #39 is retired — ADR 0015). The `architect` is always
-> installed but invoked **on-demand** —
-> outside the linear flow (P4 slice 2 authored it + its skills). `ui-designer` stays
-> inactive (#11).
-
-These are the **vendor role catalog templates**, not client instances. The
-installer renders them into a client's `.claude/agents/<role>.md` with the repo's
-resolved skills (decision #31, #32).
-
-The harness runs as **1 hat + 4 sub-agents** plus one inactive slot (decisions #10,
-#29):
-
-| Role                              | Reification                          | When invoked                                       |
-| --------------------------------- | ------------------------------------ | -------------------------------------------------- |
-| [orchestrator](./orchestrator.md) | hat (continuous, human-facing)       | entry point, RESUME, closeout, discovery mediation |
-| [spec-author](./spec-author.md)   | sub-agent (fresh context)            | PRD → spec, create GitHub issue                    |
-| [implementer](./implementer.md)   | sub-agent (fresh context)            | implement spec via TDD                             |
-| [reviewer](./reviewer.md)         | sub-agent (fresh context)            | validate against spec, tests, security             |
-| [architect](./architect.md)       | sub-agent, on-demand                 | ADRs, architecture docs, playbook iteration        |
-| [ui-designer](./ui-designer.md)   | slot — **inactive by default** (#11) | —                                                  |
-
-**Fresh context** is the distinctive value of a sub-agent over the hat: it prevents
-confirmation bias and forces self-sufficient artifacts.

package/catalog/hooks/README.md

@@ -1,56 +0,0 @@
-# hooks/ — lifecycle & enforcement hooks
-
-Shell hooks the installer wires into the client's `.claude/settings.json` and
-copies executable under `.claude/hooks/`. Helpers shared between hooks live in
-`catalog/hooks/lib/` and are copied alongside.
-
-## Hooks
-
-| Hook                  | Trigger                    | Purpose                                                                                                                                                                                                                                        |
-| --------------------- | -------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `init.sh`             | `SessionStart`             | Orient the session (read-only state summary). Blocks on hard errors (config invalid, state corruption, no git user, no `gh` auth), warns otherwise (version drift, peer hook missing, branch behind, residual markers). <1s p99. Decision #54. |
-| `session-close.sh`    | `SessionEnd` + `/pause`    | Emit `session_closed` event with duration math, write `current-<user>.md` / `sessions/<user>/{ts}.md`. No auto-commit (prints `git status --porcelain`). Decision #52.                                                                         |
-| `require-playbook.sh` | `PreToolUse` (Write\|Edit) | Block writes that match a playbook's `applies_to` glob list until that playbook has been read in this conversation. Decision #11. See `catalog/playbook-format.md`.                                                                            |
-| `suggest-playbook.sh` | `UserPromptSubmit`         | Non-blocking nudge — emit a `<system-reminder>` when a prompt matches a playbook's `keywords` regex list and that playbook hasn't been read yet. Decision #11.                                                                                 |
-
-## Shared lib
-
-`lib/transcript-grep.sh` — find evidence in the JSONL transcript (parent +
-`<transcript>/subagents/agent-*.jsonl`) that Claude actually called the `Read`
-tool with a given path. Required on day 1 by both `require-playbook.sh` and
-`suggest-playbook.sh` to avoid a misfire whereby sub-agent reads (Implementer,
-Reviewer always run as sub-agents) would not satisfy the parent's gate.
-
-`lib/playbook-scan.sh` — discover playbooks across the two layers
-(`docs/playbooks/<topic>.md` local; `~/.claude/playbooks/<topic>.md` global; local
-shadows global) and expose `playbook_scan_for_path` + `playbook_scan_for_prompt`.
-Reads frontmatter via a single `awk` pass (no `yq` — decision #29) that
-supports the documented format subset — see `catalog/playbook-format.md`.
-
-## Fail-open philosophy
-
-The three enforcement hooks (`session-close.sh`, `require-playbook.sh`,
-`suggest-playbook.sh`) fail **open**: when their dependencies (`jq`, the
-transcript file) are missing or malformed, they exit 0 — possibly with a
-stderr warning — rather than blocking the agent. They are an enforcement
-layer, not the gate of last resort.
-
-`init.sh` is the only hook that **blocks** (exit 2) — and only on hard errors
-the session cannot proceed past (missing/invalid config, state corruption,
-missing git user, missing `gh` auth). That's by design (decision #54): the
-session-start orient is the right place to fail loud, because everything
-downstream depends on those preconditions.
-
-## Bash 3.2 compatibility
-
-macOS still ships `/bin/bash` 3.2 by default. The hooks avoid `declare -A`,
-`globstar` (`**`), and other bash 4-only features so they run end-to-end on a
-stock macOS install. `init.sh`'s doctor still recommends `brew install bash` —
-the suggestion is forward-looking, not a hard dependency for these hooks.
-
-## Local testing
-
-Each hook is exercised by a Vitest spec that spawns `bash <hook>` against a
-tmpdir fixture: `src/install/init-hook.spec.ts`,
-`src/install/session-close-hook.spec.ts`, `src/install/require-playbook-hook.spec.ts`,
-`src/install/suggest-playbook-hook.spec.ts`. Run with `node --run test`.

package/catalog/playbook-format.md

@@ -1,198 +0,0 @@
-# Playbook format — specification
-
-The canonical structure for a **playbook** (the renamed _blueprint_): a project-agnostic
-document describing **how to build** a category of software (a backend service, an SPA, a
-testing strategy, a release flow). Playbooks live in the **client's** repo (and an optional
-global layer); the vendor ships **no concrete playbooks** (decision #8 — client-owned).
-
-The two enforcement hooks shipped in P5 slice 2 (`require-playbook.sh`,
-`suggest-playbook.sh`) read this format. Skills that consult playbooks read it too.
-
----
-
-## File layout
-
-A playbook is a markdown file with leading YAML frontmatter, located by **topic
-filename** across two layers:
-
-| Layer           | Path                             | Notes                    |
-| --------------- | -------------------------------- | ------------------------ |
-| Project (local) | `docs/playbooks/<topic>.md`      | Committed with the repo. |
-| Global          | `~/.claude/playbooks/<topic>.md` | Personal, generic.       |
-
-The **filename stem is the topic** — there is no index. Lookup is **local → global**: if
-both layers define `<topic>.md`, the **local one wins** and the global one is ignored for
-that topic. A missing playbook is **not an error** — agents that consult one degrade
-gracefully when it is absent.
-
-> Both layers are **configurable** via `harness.config.yml`: set `paths.playbooks`
-> (repo-relative) to relocate the local layer, and `paths.playbooks_global`
-> (absolute or `~`-based) to relocate the global one. The hooks read those keys
-> directly on each fire; editing them takes effect on the next fire (no install
-> or restart). Omit a key to keep the default shown above.
-
----
-
-## Frontmatter
-
-```yaml
----
-name: testing
-applies_to:
-  - '**/*.spec.ts'
-  - '**/*.spec.tsx'
-  - '**/*.bench.ts'
-keywords:
-  - 'vitest'
-  - 'test runner'
-  - 'AAA pattern'
-status: active
----
-```
-
-### Fields
-
-| Field        | Required | Type                            | Purpose                                                                                                    |
-| ------------ | -------- | ------------------------------- | ---------------------------------------------------------------------------------------------------------- |
-| `name`       | yes      | kebab-case string               | Identifier. Must equal the filename stem (`testing.md` → `name: testing`).                                 |
-| `applies_to` | no       | list of bash globs              | Files whose `Write`/`Edit` should be **blocked** until this playbook is read. Empty/absent → never blocks. |
-| `keywords`   | no       | list of regex strings           | Prompt patterns that **suggest** reading this playbook (non-blocking). Empty/absent → never suggests.      |
-| `status`     | no       | `draft \| active \| deprecated` | Author intent. Hooks ignore status today; reserved for skills/tooling.                                     |
-
-`applies_to` and `keywords` are **independently optional**: a playbook can be require-only
-(globs but no keywords), suggest-only (keywords but no globs), both, or neither (a doc
-agents read on demand but is not enforced).
-
-> **Frontmatter subset (awk-parsed).** The hooks read frontmatter with a single
-> `awk` pass over every discovered playbook — no `yq` (decision #29: one `yq`
-> fork per playbook blew the <50ms p99 budget, and its only edge here — YAML
-> escape-sequence fidelity — is unused because keywords are constrained to the
-> awk subset anyway). Supported: block lists (`field:\n  - item`) and simple
-> inline lists (`field: [a, b]`) of scalar strings, optionally single- or
-> double-quoted (quotes are stripped). The reader does **not** unescape YAML double-quote
-> sequences (`"\b"` stays the two characters `\b`, not a backspace) and does
-> **not** respect quotes when splitting an inline list on commas — a
-> `keywords: ["foo, bar"]` splits into `"foo` and `bar"`. Author keywords
-> within this subset: no `\b` word boundaries, no comma-bearing strings.
-> (`yq` has since left the harness entirely — decision #41 moved `init.sh` config
-> validation and `session-close.sh` state to pure bash, so `jq` is the only hard
-> dependency.)
-
-### `applies_to` semantics
-
-Each pattern is a glob-style string, transliterated to a POSIX-ERE regex and
-matched against the **full file path passed to Write/Edit** with bash
-`[[ =~ ]]`. The regex path (rather than `shopt -s globstar`) keeps the hooks
-runnable on stock macOS `/bin/bash` 3.2. Supported subset:
-
-| Glob      | Means                                                                                                                                                                                        |
-| --------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| `*`       | any sequence within a single path component (no `/`)                                                                                                                                         |
-| `**/`     | zero or more leading path components (so `**/x` matches `x` at root too)                                                                                                                     |
-| `**`      | any sequence including `/`                                                                                                                                                                   |
-| `?`       | any single non-separator character                                                                                                                                                           |
-| `[abc]`   | character class (passthrough)                                                                                                                                                                |
-| `[!abc]`  | negated class (translated to ERE `[^abc]`)                                                                                                                                                   |
-| `{a,b,c}` | alternation, becomes `(a\|b\|c)`. Nested braces and **empty alternatives** (`{,X}`, `{X,}`, `{,}`) are not supported — BSD ERE rejects empty branches; the playbook silently fails to match. |
-| `\X`      | literal `X` (escape)                                                                                                                                                                         |
-
-Examples (with the absolute path Claude Code passes in mind):
-
-| Pattern                             | Matches                                                         |
-| ----------------------------------- | --------------------------------------------------------------- |
-| `**/*.spec.ts`                      | any `*.spec.ts` at any depth, including root                    |
-| `**/apps/web/**/*.tsx`              | TSX files under any `apps/web/`                                 |
-| `**/Dockerfile`                     | a Dockerfile at any depth, including root                       |
-| `**/Dockerfile*`                    | `Dockerfile` plus variants (`Dockerfile.web`, `Dockerfile.dev`) |
-| `**/.github/workflows/*.{yml,yaml}` | CI workflow files                                               |
-
-Tip: prefix patterns with `**/` rather than relying on a bare filename — Claude
-Code passes **absolute** file paths to `Write`/`Edit`, so a literal `Dockerfile`
-(without `**/`) only matches a target whose path is exactly that string.
-
-Order **within a playbook** does not matter — any pattern matching the file path triggers
-the require. Order **across playbooks** does not matter either — if two playbooks both
-match the same path, the agent must read **both** before the write passes.
-
-**Chicken-and-egg:** a playbook whose `applies_to` matches its own location —
-e.g. `**/*.md` — will block the very file you're trying to author. Prefer
-narrow globs (file-type-specific), and if you ever need a generic-Markdown
-playbook, carve out `docs/playbooks/**` from the pattern.
-
-### `keywords` semantics
-
-Each entry is a **case-insensitive regex** evaluated against the raw user prompt
-(multi-line aware — the matcher slurps the prompt into a single string before
-testing) with `jq`'s Oniguruma engine (PCRE-like). Use `\b` for word
-boundaries, `|` for alternation inside a single pattern, etc.
-
-Examples:
-
-| Pattern                    | Fires on                               |
-| -------------------------- | -------------------------------------- |
-| `\bvitest\b`               | "let's add vitest", "Vitest config"    |
-| `\b(new pod\|nuevo pod)\b` | English or Spanish                     |
-| `\bMongo(DB)? aggregate\b` | "MongoDB aggregate", "Mongo aggregate" |
-
-Multiple playbooks may match the same prompt — all matching playbooks are suggested in
-a single `<system-reminder>` block.
-
----
-
-## How the hooks interpret it
-
-### `require-playbook.sh` (PreToolUse on `Write`/`Edit`)
-
-1. Scan all playbooks: `docs/playbooks/*.md` (local) + `~/.claude/playbooks/*.md` (global,
-   only for topics not shadowed by a local file).
-2. For each playbook with non-empty `applies_to`, test every glob against the target
-   `file_path`. If any matches, the playbook is **required**.
-3. For each required playbook, grep the conversation transcript (and any sub-agent
-   transcripts under `<transcript>/subagents/*.jsonl`) for evidence that **Claude actually
-   read it** via the `Read` tool with the playbook's absolute path.
-4. If at least one required playbook is unread, exit 2 with a message naming the missing
-   playbook(s) and the exact `Read(...)` call the agent must make. Otherwise exit 0.
-
-### `suggest-playbook.sh` (UserPromptSubmit, non-blocking)
-
-1. Scan all playbooks (same precedence as above).
-2. For each playbook with non-empty `keywords`, test every regex against the prompt. If
-   any matches, the playbook is **suggested**.
-3. Drop playbooks already read in this conversation (transcript grep, same as above).
-4. If any remain, emit a `<system-reminder>` listing them. Always exit 0.
-
-### Fail-open philosophy
-
-Both **playbook** hooks (`require-playbook.sh`, `suggest-playbook.sh`) fail
-**open** — when `yq` is missing, the transcript is unreadable, or a playbook
-frontmatter cannot be parsed, they exit 0 (suggest) or 0 with a warning
-(require) rather than blocking the agent.
-
-The lifecycle hook `init.sh` (`SessionStart`) is the one exception: it
-**blocks** by design (decision #54) on hard errors that the session cannot
-proceed past — missing/invalid config, state corruption, missing git user,
-missing `gh` auth. See `catalog/hooks/README.md` for the full fail-open vs
-blocking matrix.
-
-### No cache
-
-Each hook fire scans the catalog fresh. Playbooks are read sparingly (PreToolUse fires
-only on Write/Edit; UserPromptSubmit fires once per turn) and the catalog is small. A
-cache would add a staleness failure mode without measurably improving latency.
-
----
-
-## Authoring conventions
-
-- One topic per file. Split when a topic grows beyond a single screen of frontmatter.
-- Project-**specific** rules — a business rule, a quirk of one app — do **not** belong
-  here. Put them in `CLAUDE.md`, a project-local ADR (`docs/adr/`), or `CONTEXT.md`.
-- Reference other playbooks with `[topic](./topic.md)` — relative links so they resolve
-  in any Markdown viewer.
-
-### Iteration
-
-Playbooks evolve with the codebase. The **Architect**'s `playbook-iterate` skill proposes
-edits when a spec collides with an existing playbook (a `T6 PLAYBOOK_CONFLICT` review
-finding) or when a recurring pattern in the code is not yet captured. The Architect
-proposes; the client decides.

package/catalog/schemas/README.md

@@ -1,13 +0,0 @@
-# schemas/ — telemetry & config schemas
-
-Schema definitions the harness writes/validates against. Tier 1 (client-local)
-writes events in these formats from day one so the data is forward-compatible with
-the Tier 2 central backend designed in Fase 1+.
-
-## Documents
-
-| File                                                 | Decision | Status                     | Purpose                                                                                                                                                                                                                                                                                                                                                                       |
-| ---------------------------------------------------- | -------- | -------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| [`tier2-events.md`](tier2-events.md)                 | #27, #61 | **Authority — P5.**        | Event schema for `events.jsonl`: 9 event types (`session_closed`, `spec_created`, `spec_approved`, `task_done`, `review_rejected`, `bug_post_merge`, `l3_bypass`, `followup_captured`, `step_completed`), the common envelope (`type`, `ts`, `user`, `project`, `task_id`, `harness_version`), and per-field tags (`sensitive`/`internal`/`metric`) for forward-sanitization. |
-| [`tier2-events-history.md`](tier2-events-history.md) | #51      | Seeded at `0.1.0-alpha.0`. | Forward-only per-release deltas (ADDED/REMOVED/DEPRECATED/RENAMED fields, new event types) — supports forward-only, dispatch-on-read. ~30 min/release.                                                                                                                                                                                                                        |
-| `harness.config.schema.json`                         | #53(e)   | Pending (S3).              | JSON Schema generated from the Zod config schema via `zod-to-json-schema`, distributed for IDE autocomplete.                                                                                                                                                                                                                                                                  |

package/catalog/skills/README.md

@@ -1,62 +0,0 @@
-# skills/ — vendor skill catalog
-
-> **Status: 18 skills (P1–P3 + P4).** P1 migrated `triage-issue`, `tdd`,
-> `senior-review`. P2 migrated `grill-with-docs` and authored `prd-to-spec`,
-> `spec-to-issue`, `task-closeout`. P3 authored the discovery loop (`raise-discovery`,
-> `resolve-discovery`). **P4 slice 1** added the Reviewer set — `verify`,
-> `silent-failure-hunter`, `security-review`, `spec-compliance-check`,
-> `test-gap-report` — rewrote `senior-review` to v2, and rolled out the gating
-> frontmatter (`min-profile`/`phase`/`invoked-by`) across the catalog. **P4 slice 2**
-> added the Architect set — `write-adr`, `update-architecture`, `code-explorer`,
-> `playbook-iterate`. All `origin: vendor`.
-
-Generic, **project-agnostic** skills only (decision #28). Project-specific skills
-(e2e, changeset, docs) live in the **client's** `.claude/skills/`, not here. Each
-skill is a folder `skills/<name>/SKILL.md`.
-
-## Gating frontmatter (decision #31; ADR 0015)
-
-The installer scans this catalog, parses each skill's frontmatter, and lands a skill
-when every **`applies-when`** capability key holds for the repo (install-time,
-deterministic). A skill with no `applies-when` is always installed — there is no
-profile tier (the coarse `min-profile` filter #39 was retired in ADR 0015, when
-profiles collapsed to a single capability-gated skill set). Each sub-agent's
-`{{SKILLS}}` marker is filled with the skills it `invoked-by`, grouped by **`phase`**.
-A skill with no `phase` is universal (e.g. `raise-discovery`).
-
-| Field               | Meaning                                                                | Default          |
-| ------------------- | ---------------------------------------------------------------------- | ---------------- |
-| `phase`             | `pre-implementation` / `during-implementation` / `post-implementation` | none ⇒ universal |
-| `invoked-by`        | roles whose marker lists it                                            | `[]`             |
-| `applies-when`      | capability keys (AND) the repo must satisfy at install                 | `[]` (always)    |
-| `trigger-condition` | per-change runtime guard rendered with the skill                       | —                |
-
-## Planned MVP catalog (Fase 0)
-
-| Skill                   | Role                               | Status             |
-| ----------------------- | ---------------------------------- | ------------------ |
-| `grill-with-docs`       | Orchestrator (shared w/ Architect) | migrated ✓ (P2)    |
-| `triage-issue`          | Orchestrator                       | migrated ✓ (P1)    |
-| `task-closeout`         | Orchestrator                       | authored ✓ (P2)    |
-| `prd-to-spec`           | Spec Author                        | authored ✓ (P2)    |
-| `spec-to-issue`         | Spec Author                        | authored ✓ (P2)    |
-| `tdd`                   | Implementer                        | migrated ✓ (P1)    |
-| `senior-review`         | Reviewer                           | v2 ✓ (P4 s1)       |
-| `verify`                | Implementer, Reviewer              | authored ✓ (P4 s1) |
-| `security-review`       | Reviewer                           | authored ✓ (P4 s1) |
-| `silent-failure-hunter` | Reviewer                           | authored ✓ (P4 s1) |
-| `spec-compliance-check` | Reviewer                           | authored ✓ (P4 s1) |
-| `test-gap-report`       | Reviewer                           | authored ✓ (P4 s1) |
-| `write-adr`             | Architect                          | authored ✓ (P4 s2) |
-| `update-architecture`   | Architect                          | authored ✓ (P4 s2) |
-| `code-explorer`         | Architect                          | authored ✓ (P4 s2) |
-| `playbook-iterate`      | Architect                          | authored ✓ (P4 s2) |
-| `raise-discovery`       | all sub-agents (universal)         | authored ✓ (P3)    |
-| `resolve-discovery`     | Orchestrator                       | authored ✓ (P3)    |
-
-**Parked from the vendor MVP:** `feature-flow`, `prd-to-plan`, `prd-to-issues`,
-`pr-review`, `ui-design`, `project-setup`, `write-a-skill`, `grill-me` (deprecated).
-
-**ECC as marketplace (decision #42):** [`affaan-m/ECC`](https://github.com/affaan-m/ECC)
-is MIT; when a client needs a specific skill ECC already has, copy-and-adapt it into
-the client's `.claude/skills/` rather than bloating this catalog.

package/catalog/templates/README.md

@@ -1,32 +0,0 @@
-# templates/ — installer render sources
-
-> **Status: claude-code templates (P1–P4 s2).** `agents.md.tpl`,
-> `harness.config.yml.tpl`, `state/history.md.tpl`, and `docs/playbooks/README.md.tpl`
-> (the playbooks-home convention doc, P4 s2) exist and are rendered by the installer.
-> `.claude/commands/`, `.claude/settings.json`, and the rest of `docs/` land in later
-> phases.
-
-Files the installer **renders into a client repo** during `install`. Templates use
-placeholders (e.g. `{{vendor_version}}`, `{{target}}`, and the per-agent `{{SKILLS}}`
-marker) resolved from the repo capability scan. They are organized by `target`
-(decision #38), so additional targets plug in without a refactor.
-
-## Planned layout
-
-```
-templates/
-└── claude-code/                 # only operative target in Sprint 0
-    ├── agents.md.tpl            # entry protocol / dispatcher
-    ├── harness.config.yml.tpl   # config with {{capabilities}} placeholders
-    ├── .claude/
-    │   ├── agents/<role>.md.tpl  # role instances (PRE/DURING/POST filled)
-    │   ├── commands/*.md.tpl     # /resume, /define, /triage, /pause, /bypass, /hotfix
-    │   └── settings.json.tpl     # hooks, permissions
-    ├── state/
-    │   ├── current.md.tpl
-    │   └── history.md.tpl
-    └── docs/
-        ├── playbooks/README.md.tpl  # playbooks-home convention (P4 s2)
-        ├── CONTEXT.md.tpl
-        └── adr/.tpl
-```