@event4u/agent-config 9.8.0 → 9.9.0

package/docs/architecture/setup-vs-settings-shared-surface.md

@@ -1,114 +0,0 @@
-# Shared Surface — Setup Wizard ↔ Settings Editor
-
-> Single source of truth for the "wizard and settings editor live in one
-> codebase" promise of
-> [`agents/roadmaps/unified-setup-and-settings-gui.md`](../../agents/roadmaps/unified-setup-and-settings-gui.md).
-> Folded in during Phase 0 of implementation from the
-> `frontend-engineer` council finding (line 332 of the roadmap).
-
-## Rule
-
-Every UI component, validator, schema, and API surface in
-`src/ui/` and `src/server/` is **one of three** labels:
-
-| Label | Meaning | Drift policy |
-|---|---|---|
-| `shared` | Used by **both** the wizard and the settings editor, unmodified | Any divergence is a regression |
-| `setup-only` | Wizard chrome — meaningless in the settings editor | Settings-page must never import |
-| `settings-only` | Advanced fields the wizard hides for first-run brevity | Wizard must never expose |
-
-End of every phase, the implementing agent walks `src/ui/` + `src/server/`
-and verifies each new file matches one of the three. Phase review fails
-on drift.
-
-## Inventory (Phase 0 baseline)
-
-### Schema + validation
-
-| Module | Label | Notes |
-|---|---|---|
-| `src/server/schemas/settings.ts` | `shared` | Wizard ⊂ settings — wizard slices, never owns its own schema |
-| `src/server/schemas/userMd.ts` | `shared` | Parser convergence locked (gray-matter, ADR-010) |
-| `src/server/schemas/wizardState.ts` | `setup-only` | Partial-state-on-disk shape |
-
-### Server routes
-
-| Route | Label |
-|---|---|
-| `GET/PUT /api/v1/settings` | `shared` |
-| `GET /api/v1/settings/diff` | `shared` |
-| `GET/PUT /api/v1/user-md` | `shared` |
-| `GET /api/v1/user-md/template` | `shared` |
-| `GET /api/v1/wizard/state` | `setup-only` |
-| `POST /api/v1/wizard/state` | `setup-only` |
-| `POST /api/v1/wizard/finish` | `setup-only` |
-
-### IO primitives
-
-| Module | Label |
-|---|---|
-| `src/server/io/atomicWrite.ts` | `shared` |
-| `src/server/io/atomicMultiWrite.ts` | `shared` |
-| `src/server/io/substituteTemplate.ts` | `shared` |
-
-### UI form primitives (Phase 2)
-
-| Module | Label |
-|---|---|
-| `src/ui/forms/Field.tsx` | `shared` |
-| `src/ui/forms/TextInput.tsx` | `shared` |
-| `src/ui/forms/NumberInput.tsx` | `shared` |
-| `src/ui/forms/Toggle.tsx` | `shared` |
-| `src/ui/forms/Radio.tsx` | `shared` |
-| `src/ui/forms/Textarea.tsx` | `shared` |
-| `src/ui/forms/Autocomplete.tsx` | `shared` |
-| `src/ui/forms/FieldError.tsx` | `shared` |
-| `src/ui/forms/FieldDescription.tsx` | `shared` |
-| `src/ui/forms/SchemaForm.tsx` | `shared` |
-| `src/ui/copy/errors.ts` | `shared` |
-
-### UI pages (Phase 2)
-
-| Module | Label |
-|---|---|
-| `src/ui/pages/SettingsPage.tsx` | `settings-only` |
-| `src/ui/pages/UserMdPanel.tsx` | `shared` (the wizard's UserMdStep wraps this body editor) |
-| `src/ui/pages/DiffModal.tsx` | `shared` |
-
-### UI wizard chrome (Phase 3)
-
-| Module | Label |
-|---|---|
-| `src/ui/wizard/StepHeader.tsx` | `setup-only` |
-| `src/ui/wizard/ProgressBar.tsx` | `setup-only` |
-| `src/ui/wizard/StepNav.tsx` | `setup-only` |
-| `src/ui/wizard/FinalActions.tsx` | `setup-only` |
-| `src/ui/wizard/steps/*.tsx` | `setup-only` (each step composes `shared` form primitives) |
-| `src/ui/wizard/WizardPage.tsx` | `setup-only` |
-
-### CLI
-
-| Module | Label |
-|---|---|
-| `src/cli/commands/uiServe.ts` | `shared` |
-| `src/cli/commands/settings.ts` | `settings-only` (initial-route flag only) |
-
-## Audit procedure
-
-End of every UI / server PR, run:
-
-```bash
-node scripts/check-shared-surface.mjs  # walks src/ui + src/server, prints unlabeled files
-```
-
-If the script is not yet written (it lands in Phase 4 follow-up), the
-reviewer audits this table by eye against the diff. Unlabeled new
-modules fail review.
-
-## Why this matters
-
-The whole roadmap is justified by reusing one codebase across two
-surfaces. Without this table, every PR can quietly diverge — wizard
-form fields drift from settings fields, primitives get forked, the
-shared-codebase claim erodes. The table is the contract; the audit
-is the enforcement.

package/docs/architecture/source-projection.md

@@ -1,85 +0,0 @@
-# Pipeline A — Source projection
-
-> **Scope:** transform verbose authoring source into the deterministic
-> distribution payload that ships in the npm package. The pipeline does
-> path-rewriting, `.npmignore`-style filtering, hash-tracking, and (on
-> selected files) telegraph-style prose condensation. The *primary* function
-> is the source-to-dist projection itself; raw byte reduction is small
-> (~0.35 % on the source/dist boundary: 3,253,997 B → 3,242,579 B across
-> 596 files) because most files are 1:1-projected with only frontmatter
-> and link rewrites. Per-tool size at the downstream projection
-> boundaries (`.augment/`, `.claude/`, `.cursor/`, `.windsurf/`,
-> `.clinerules/`, `.windsurfrules`, `GEMINI.md`) is measured separately
-> — see [`multi-tool-projection.md § Per-tool projection size`](multi-tool-projection.md#per-tool-projection-size)
-> for the table produced by [`scripts/measure_projection_bytes.py`](../../src/scripts/measure_projection_bytes.py).
-
-> **Historical note.** This pipeline was previously labelled
-> "Condensation". Renamed in the v2.10.0 feedback follow-up after the
-> council pointed out that the dominant function is projection, not
-> byte condensation. The script names (`scripts/condense.py`,
-> `scripts/condense.sh`) are kept for now to avoid a large blast-radius
-> refactor; the prose-condensation sub-step (and the `/condense` slash
-> command for telegraph text condensation) still earn the legacy name.
-
-## Input → Transform → Output
-
-```
-.agent-src.uncondensed/**         ← Source of truth (verbose, human-readable)
-    ↓ scripts/condense.py + scripts/condense.sh (--sync)
-dist/agent-src/**                      ← Condensed, hash-tracked, shipped in @event4u/agent-config
-```
-
-| Layer | Source | Output |
-|---|---|---|
-| Skills | `.agent-src.uncondensed/skills/<id>/SKILL.md` (+ assets) | `dist/agent-src/skills/<id>/SKILL.md` |
-| Rules | `.agent-src.uncondensed/rules/<name>.md` | `dist/agent-src/rules/<name>.md` |
-| Commands | `.agent-src.uncondensed/commands/**` | `dist/agent-src/commands/**` |
-| Personas, contexts, templates | `.agent-src.uncondensed/<dir>/**` | `dist/agent-src/<dir>/**` |
-
-The path rewriter ([`scripts/condense.py:157`](../../src/scripts/condense.py)
-`apply_path_rewriter()`) converts logical names in source frontmatter
-(`contexts/execution/foo.md`) into the relative form expected from
-`dist/agent-src/rules/` (`../contexts/execution/foo.md`). Hardcoding
-`.agent-src.uncondensed/` in source is a CI failure — caught by
-[`scripts/check_condensed_paths.py`](../../src/scripts/check_condensed_paths.py).
-
-## Entry points
-
-| Surface | Command |
-|---|---|
-| Full sync | `task sync` ([`taskfiles/content.yml:4`](../../taskfiles/content.yml)) |
-| Sync drift check | `task sync-check` ([`taskfiles/content.yml:38`](../../taskfiles/content.yml)) |
-| Hash drift check | `task sync-check-hashes` ([`taskfiles/content.yml:43`](../../taskfiles/content.yml)) |
-| Single file | `task sync-mark-done -- <path>` |
-| Direct script | `bash scripts/condense.sh --sync` |
-
-## Invariants
-
-1. **Determinism** — same input must produce identical bytes in
-   `dist/agent-src/`. CI enforces via `task sync-check` (no output diff
-   permitted on a clean checkout).
-2. **Hash tracking** — every condensed file's source-hash is stored
-   in `dist/agent-src/.condensation-hashes.json`; stale hashes are caught
-   by `task sync-check-hashes`.
-3. **No source-side leakage** — `.agent-src.uncondensed/` must not
-   appear anywhere in condensed output (frontmatter, body, includes).
-4. **Frontmatter preserved** — YAML frontmatter survives condensation
-   unchanged except for the path rewriter on `load_context:` entries.
-
-## Failure modes
-
-| Symptom | Cause | Fix |
-|---|---|---|
-| `task sync-check` fails on clean tree | source edited but not re-condensed | `task sync` |
-| `check-condensed-paths` fails | `.agent-src.uncondensed/` substring leaked into condensed output | re-author source, re-run `task sync` |
-| Hash drift on unchanged file | concurrent edits / merge artefact | `task sync-clean-hashes && task sync` |
-| Path rewriter mangles a link | logical name collision with a real relative path | declare `validator_ignore:` in rule frontmatter |
-
-## Proving the pipeline
-
-- [`tests/test_condense.py`](../../tests/test_condense.py) — end-to-end
-  condensation, hash invariants, path rewriter.
-- [`tests/test_condense_paths.py`](../../tests/test_condense_paths.py)
-  — path-rewriter edge cases and forbidden-substring detection.
-
-← [Architecture overview](../architecture.md)

package/docs/architecture.md

@@ -1,390 +0,0 @@
-# Architecture
-
-> **agent-config is not a runtime, but it ships a deterministic orchestration contract / state machine for host agents.**
-
-## System overview
-
-Six layers, ordered from "how the package reaches a consumer" down to "what a consumer's agent actually executes". Each layer names its canonical contract under [`docs/contracts/`](contracts/) — the overview is a router, not a re-statement.
-
-```
-Distribution         → npx-only runtime · install.sh · lockfile pin        ← stable
-Governance           → Kernel rules · tier-1/2 routing · command clusters  ← stable
-Router-Kernel        → dist/router.json · always-loaded Iron Laws · char caps   ← stable
-Projection           → Source · augment / multi-tool / cloud bundles       ← stable
-Execution Contracts  → Skills · commands · work-engine · roadmap engine    ← stable / beta
-MCP Lite/Full        → Hosted read-only (Lite) · local stdio (Full)        ← experimental
-```
-
-| Layer | Canonical contract | Tier |
-|---|---|---|
-| **Distribution** | [`installed-tools-lockfile.md`](contracts/installed-tools-lockfile.md) + the "Distribution model" subsection below | (beta) |
-| **Governance** | [`command-clusters.md`](contracts/command-clusters.md) + [`command-surface-tiers.md`](contracts/command-surface-tiers.md) | (beta) |
-| **Router-Kernel** | [`kernel-membership.md`](contracts/kernel-membership.md) + [`rule-router.md`](contracts/rule-router.md) | (beta) |
-| **Projection** | [`architecture/source-projection.md`](architecture/source-projection.md), [`augment-projection.md`](architecture/augment-projection.md), [`multi-tool-projection.md`](architecture/multi-tool-projection.md), [`claude-bundle.md`](architecture/claude-bundle.md) | stable |
-| **Execution Contracts** | [`implement-ticket-flow.md`](contracts/implement-ticket-flow.md), [`orchestration-dsl-v1.md`](contracts/orchestration-dsl-v1.md), [`adr-product-ui-track.md`](contracts/adr-product-ui-track.md) | stable (adr-product-ui-track) / (beta) (implement-ticket-flow · orchestration-dsl-v1) |
-| **MCP Lite/Full** | `docs/contracts/mcp-phase-1-scope.md`, `docs/contracts/mcp-cloud-scope.md`, `docs/contracts/mcp-beta-criteria.md` | experimental — promotion to beta gated on `mcp-beta-criteria.md` (six artefact gates, monitored by `agent-config doctor --check mcp-beta-readiness`) |
-
-Stability tiers follow [`docs/contracts/STABILITY.md`](contracts/STABILITY.md):
-
-- **stable** = shipped, documented, exercised by the default (`minimal`) profile or by CI on every PR; SemVer-major for breaks.
-- **beta** = shipped and load-bearing for one or more flows, but the surface is expected to evolve; minor-version breaks allowed under a `### Breaking` CHANGELOG note.
-- **experimental** = scaffold or pilot status; breaks allowed in any release.
-
-### What changed since 2.2.2
-
-Four load-bearing additions reshaped the top of the model between 2.2.2 and the current release. They are listed here so the diagram above reads as the *current* package, not a historical accumulation:
-
-1. **Router-Kernel** — the always-loaded Iron Laws collapsed into a 9-rule kernel with explicit per-rule character budgets enforced by `task lint-rule-budget`; everything else routes via tier-1/2 (`dist/agent-src/router.json`). Contract: [`kernel-membership.md`](contracts/kernel-membership.md) + [`rule-router.md`](contracts/rule-router.md) — both (beta).
-2. **MCP Lite/Full** — replaces the old "Tool Adapters" layer at the top level. Lite is the hosted read-only surface (Claude.ai, Cloud agents); Full is the local stdio server consumers self-host. Promotion to beta is gated on six falsifiable artefacts in `docs/contracts/mcp-beta-criteria.md`; the old GitHub / Jira adapters remain as an internal detail of the Execution Contracts layer (see Tool Adapters subsection below).
-3. **npx distribution** — Composer and `npm install` paths retired in favour of `npx @event4u/agent-config`, with the lockfile-equivalent role played by `agent_config_version` in `.agent-settings.yml`. Full rationale in the "Distribution model" subsection below.
-4. **Command tiering** — `/`-commands now declare a `tier:` (0 / 1 / 2 / 3) that maps to invocation frequency and surface budget; tier-0 is the trimmed Tier-0 set surfaced in `agent-config --help` after the 2.7.x surface-discipline pass. Contract: [`command-surface-tiers.md`](contracts/command-surface-tiers.md) + [`command-clusters.md`](contracts/command-clusters.md) — both (beta).
-5. **TypeScript CLI shell** — the `agent-config` binary is now a compiled TS entry point (`dist/cli/agent-config.js`) that handles native commands (`versions`, `doctor-shell`, `ui:serve`) and forwards every other subcommand to the legacy Bash dispatcher at `scripts/_dispatch.bash`. The shell also embeds a Fastify-based local server bound to `127.0.0.1` for the in-browser UI track. Decision: [`docs/decisions/ADR-012-typescript-cli-shell.md`](decisions/ADR-012-typescript-cli-shell.md). Wire shape + security stance: [`local-server-api.md`](contracts/local-server-api.md) (beta).
-
-> The previous "observability, feedback, lifecycle" layers were removed in
-> 1.5 — they were scaffolds without production consumers. The "Tool
-> Adapters" top-level layer was demoted in 2.7 — see point 2 above.
-
-## Content pipelines
-
-The end-to-end flow from source authoring to distributed surfaces
-is four discrete pipelines, each owned by one sub-page. Each page
-documents input → transform → output, invariants, failure modes,
-and cites the script, Taskfile target, and test file that prove
-the pipeline.
-
-```
-src/                      ──Pipeline A──▶  dist/agent-src/
-                                              │
-                                              ├──Pipeline B──▶  .augment/
-                                              │
-                                              ├──Pipeline C──▶  .claude/ · .cursor/ · .clinerules/
-                                              │                  .windsurfrules · GEMINI.md
-                                              │
-                                              └──Pipeline D──▶  dist/cloud/<skill>.zip
-```
-
-| Pipeline | Page | Output |
-|---|---|---|
-| **A.** Source projection | [`architecture/source-projection.md`](architecture/source-projection.md) | `dist/agent-src/` |
-| **B.** Augment projection | [`architecture/augment-projection.md`](architecture/augment-projection.md) | `.augment/` |
-| **C.** Multi-tool projection | [`architecture/multi-tool-projection.md`](architecture/multi-tool-projection.md) | `.claude/`, `.cursor/`, `.clinerules/`, `.windsurfrules`, `GEMINI.md` |
-| **D.** Claude.ai bundle | [`architecture/claude-bundle.md`](architecture/claude-bundle.md) | `dist/cloud/<skill>.zip` |
-
-The drift check
-[`tests/test_architecture_docs_pipelines.py`](../tests/test_architecture_docs_pipelines.py)
-fails if any of the four sub-pages exists without its cited script /
-Taskfile target — or vice versa.
-
-### Canonical distribution channel per AI tool
-
-For every tool the package supports, **filesystem** is the canonical channel that the consumer installer (`scripts/install.sh`) writes — see [`contracts/skill-distribution-channels.md`](contracts/skill-distribution-channels.md). Pipeline C above always writes the filesystem trees (`.claude/skills/`, `.cursor/rules/`, `.clinerules/`, `.windsurfrules`, etc.) into the consumer project. Plugin manifests at the package source (`.claude-plugin/marketplace.json`, `.augment-plugin/marketplace.json`) are preserved for users who install via the host's plugin registry, but the installer does **not** project a second registry into consumer projects. Users on older harnesses that require both channels opt in with `bash scripts/install.sh --legacy-both`. Regression test: [`tests/test_canonical_distribution.py`](../tests/test_canonical_distribution.py).
-
-Cross-references inside `dist/agent-src/rules/*.md` are written
-**relative to `dist/agent-src/rules/`** (e.g. `../contexts/execution/foo.md`,
-`../docs/guidelines/agent-infra/foo.md`). Source files under
-`src/rules/` use **logical names** without a
-directory prefix (e.g. `contexts/execution/foo.md`); the
-condense-time path rewriter in `scripts/condense.py` translates
-them to the relative form when writing into `dist/agent-src/`. Hardcoding
-`.agent-src.uncondensed/` in source frontmatter or body links is
-forbidden and caught by `scripts/check_condensed_paths.py`.
-
-### Distribution model — npx-only + version-pin governance
-
-Starting with the road-to-portable-runtime cutover, the package ships
-exclusively as a **runtime resolved by `npx @event4u/agent-config`**.
-There is no Composer dependency, no `npm install` step, no `--global`
-symlink scheme. Consumers run:
-
-```bash
-npx @event4u/agent-config init      # bootstrap a project
-npx @event4u/agent-config <cmd>     # any subsequent command
-```
-
-**Why local installs are gone.** Vendoring the package into every
-consumer's `vendor/` or `node_modules/` created three problems: stale
-runtimes diverging from the published version, build-system coupling
-(Composer post-install hooks, npm postinstall scripts), and a parallel
-"global install" scheme that copied curated skills into `~/.claude/`,
-`~/.cursor/`, `~/.codeium/windsurf/` under an `event4u/` namespace.
-Maintenance cost was high, the abstractions leaked across surfaces, and
-debugging "which version is loaded" was non-trivial. `npx` resolves the
-runtime per invocation against a single npm registry source, which
-collapses all three failure modes.
-
-**How the pin replaces lockfile determinism.** A consumer-managed
-`composer.lock` / `package-lock.json` previously froze the runtime
-version per repo. Under npx-only, the equivalent role is played by the
-`agent_config_version` field in the consumer's `.agent-settings.yml`
-(see `src/config/agent-settings.template.yml`). The dispatcher reads the
-pin on every invocation and re-execs `npx @event4u/agent-config@<pin>`
-if the resolved version diverges (P3.2 pin-resolver). The pin lives in
-the consumer's repo, is reviewed in PRs, and survives `npx`'s own
-cache eviction.
-
-**Q1 council rejection + override + pin as substitute.** During Q1
-planning, the architecture council rejected the npx-only proposal on
-the grounds that `npx` resolution introduces a "09:00 vs 09:15
-release skew" window where two developers running the same command
-minutes apart could see different runtimes. The user overrode the
-rejection on the condition that an explicit version pin replaces
-lockfile determinism — the council's concern is real, but the pin
-collapses the skew window to whatever the project's PR cadence is.
-Pin drift across developers becomes a reviewable config change in
-`.agent-settings.yml` rather than an invisible registry-resolution
-race. ADR-pending entry will record the trade-off in full once P3 is
-green.
-
-### Cloud-bundle pipeline
-
-See [`architecture/claude-bundle.md`](architecture/claude-bundle.md)
-for the full Pipeline D documentation — tier classification, sandbox
-note, package-internal path-swap, description budget, and the
-`task ci-cloud-bundle` dry-run gate.
-
-## What's inside
-
-| Layer | Count | Purpose |
-|---|---|---|
-| **Skills** | 281 | On-demand expertise — stack analysis (Laravel · Symfony · Zend / Laminas · Next.js · React · Node), testing, Docker, API design, security, observability, … |
-| **Rules** | 107 | Always-active constraints — coding standards, scope control, verification, language-and-tone, agent-authority |
-| **Commands** | 190 | Slash-command workflows — `/commit`, `/create-pr`, `/fix ci`, `/optimize skills`, `/feature plan`, `/work`, `/implement-ticket`, `/condense`, … |
-| **Guidelines** | 101 | Reference material cited by skills — PHP patterns, Eloquent, Playwright, agent-infra, … |
-| **Templates** | 7 | Scaffolds for features, roadmaps, contexts, skills, overrides |
-| **Contexts** | 5 | Shared knowledge about the system itself |
-
----
-
-## Core vs full install
-
-Every shipped artefact carries a **surface tier** so experimental tooling cannot
-destabilise the lean engine adopters install:
-
-- **`core`** — the lean stable multi-host engine: rules + skills + install +
-  condensation, plus the engineering/authoring packs. This is the adoptable
-  surface. Its on-disk install layout is a frozen, versioned contract
-  ([`install-layout.md`](contracts/install-layout.md)) (beta).
-- **`lab`** — experimental / pilot meta-tooling that must not ride the same
-  release train as the core: AI media pipelines (`ai-video`, `ai-image`), the
-  `fun` pack, and the package's own pilot script clusters (council, MCP server,
-  prediction-pool, cost analytics, chat-history, benchmarking).
-
-Two install modes — the only end-user concept here:
-
-- **Full install** (default) — everything, including lab tooling. The
-  maintainer's own working surface.
-- **Core-only install** (`agent-config install --global --core-only`) — prunes
-  lab-tier skills/commands from the deployed tree, leaving a working
-  rules + skills + condensation engine with zero lab modules. The adoptable
-  surface for consumers who don't want pilot churn.
-
-A boundary guard (`scripts/check_surface_tiers.py`, in `task ci-fast`) keeps the
-two apart: it forbids a `core` module from hard-importing a `lab` module
-(guarded `try/except ModuleNotFoundError` optional imports excepted) and asserts
-the cluster-tier registry is exhaustive. Full tier map + the inventory:
-[`surface-tiers.md`](contracts/surface-tiers.md) (beta).
-
----
-
-## Execution-layer detail
-
-> The six layers in the System overview are the top-level model. This section provides depth on the **Governance**, **Router-Kernel**, and **Execution Contracts** layers — the three a host agent interacts with on every turn. Distribution and Projection live in their own sub-pages ([`architecture/`](architecture/) and the "Distribution model" subsection above); MCP Lite/Full lives in [`docs/mcp-server.md`](mcp-server.md).
-
-### 1. Governance Layer
-
-- **Rules** → always-active behavior constraints
-- **Skills** → structured, executable procedures
-- **Guidelines** → reference-only documentation
-- **Commands** → workflow orchestration
-
-Ensures: no guessing, analysis before action, real verification, consistent outputs. Canonical contracts (all (beta)): [`kernel-membership.md`](contracts/kernel-membership.md), [`rule-router.md`](contracts/rule-router.md), [`command-clusters.md`](contracts/command-clusters.md), [`command-surface-tiers.md`](contracts/command-surface-tiers.md).
-
-### 2. Runtime Dispatcher — stable mechanism, pilot coverage
-
-> **Scope:** single-skill execution. Resolves a `SKILL.md` with
-> `execution.command` argv, enforces safety constraints, hands off to
-> the matching handler. **Not** a multi-step orchestrator — that is
-> the Work Engine (next section).
-
-> **Status:**
-> - **Stable mechanism:** the dispatcher itself
->   (`scripts/runtime_dispatcher.py`), the shell handler
->   (`scripts/runtime_handler.py`), and the `ExecutionResult` shape.
->   `subprocess.run` is invoked with `shell=False` (argv only); the
->   environment is scrubbed to an explicit allowlist.
-> - **Pilot coverage:** two skills ship as live pilots —
->   `lint-skills` and `check-refs` — both run on every PR and appear
->   in the GitHub Step Summary via `scripts/ci_summary.py`.
-> - **Scaffold:** `php` and `node` handlers — the frontmatter accepts
->   them and the registry validates them, but no handler
->   implementation exists yet.
-
-Skills opt into runtime by declaring execution metadata:
-
-```yaml
-execution:
-  type: manual | assisted | automated
-  handler: shell | php | node | internal | none
-  command:                       # required for shell/php/node runtime paths
-    - ./scripts-run
-    - src/scripts/skill_linter
-    - "--all"
-  timeout_seconds: 120
-  allowed_tools: []
-  safety_mode: strict            # required for type=automated
-```
-
-Invoke a runtime-capable skill end-to-end:
-
-```bash
-./scripts-run src/scripts/runtime_dispatcher run --skill lint-skills
-```
-
-A typed `ExecutionResult` (exit code, stdout, stderr, duration,
-artifacts) is returned and can be persisted as JSON via
-`--output FILE`.
-
-Planned scope: `php` / `node` handlers, tool-registry wiring for
-`allowed_tools`, streaming output.
-
-### 3. Work Engine — beta, multi-step orchestration
-
-> **Scope:** multi-step phase dispatch for `/work` and
-> `/implement-ticket`. Drives the
-> `refine → score → plan → implement → test → verify → report` loop,
-> persists state in `.work-state.json`, and routes UI-shaped work
-> through the product UI track. Lives at
-> [`templates/scripts/work_engine/`](../src/agent-src/templates/scripts/work_engine/);
-> shipped to consumer projects via `scripts/install.py`.
-
-> **Status: beta.** The contract (directive sets, halt budgets,
-> envelope shape) has shipped one full SemVer-minor cycle, but the
-> surface is still expected to evolve. Breaks are allowed in
-> minor-version releases under a `### Breaking` CHANGELOG note. See
-> [`docs/contracts/STABILITY.md`](contracts/STABILITY.md).
-
-Key responsibilities:
-
-- **Directive routing** — `ui` / `ui-trivial` / `mixed` directive
-  sets, locked into the contract at
-  [`adr-product-ui-track.md`](contracts/adr-product-ui-track.md) (beta).
-- **Halt protocol** — every phase emits a structured halt; the
-  agent re-enters with the user's answer, never improvises.
-- **State machine** — `.work-state.json` is the single source of
-  truth across resumes; the engine refuses to switch envelope
-  mid-flight. Legacy `.implement-ticket-state.json` files are
-  detected on load and routed through
-  [`docs/MIGRATION.md`](MIGRATION.md).
-- **Hooks** — chat-history, telemetry, and platform hooks fire
-  through the engine's hook layer.
-
-The Work Engine **uses** the Runtime Dispatcher when a phase needs
-to execute a single skill (e.g. lint, refs check), but the two are
-independent components with separate stability tiers.
-
-### 4. Tool Adapters — experimental (internal detail; superseded at the top level by MCP)
-
-> **Position in the new model.** Tool Adapters no longer occupy a top-level layer — that slot is now **MCP Lite/Full**. The adapter classes still ship as the internal mechanism the Work Engine uses for inline GitHub/Jira reads, but external integration is meant to land via MCP going forward. See `docs/contracts/mcp-phase-1-scope.md`, `docs/contracts/mcp-cloud-scope.md`, and `docs/contracts/mcp-beta-criteria.md` for the surface that replaces this layer at the top level.
-
-> **Status: scaffold + read-only GitHub calls.** With a `GITHUB_TOKEN` the
-> GitHub adapter performs real read calls; without one it returns scaffold
-> data. Write operations (`create_pr`, `comment`, etc.) are scaffold only.
-> The Jira adapter is scaffold throughout.
-
-Controlled integration via adapters:
-
-- GitHub adapter (read-first: PRs, issues, files, commits)
-- Jira adapter (read-first: tickets, search)
-- Tool registry with safety rules for execution
-- Structured responses with error classification
-
-### 5. Cost Control
-
-> **Key principle:** Opt-in by default.
-
-The Runtime Dispatcher and Tool Adapters activate only under the
-`balanced` or `full` profile. The Work Engine activates whenever
-`/work` or `/implement-ticket` is invoked and is independent of the
-cost profile. The default `minimal` profile ships rules, skills, and
-commands and nothing else. All settings and their profile defaults
-are documented in
-[`src/agent-src/templates/agent-settings.md`](../src/agent-src/templates/agent-settings.md).
-
----
-
-## Supported AI Tools
-
-| Tool | Rules | Skills | Commands | Plugin | Method |
-|---|---|---|---|---|---|
-| **Augment CLI** | ✅ | ✅ | ✅ | ✅ | Native plugin (recommended) |
-| **Augment VSCode/IntelliJ** | ✅ | ✅ | ✅ | — | install.sh (copies + symlinks) |
-| **Claude Code** | ✅ | ✅ | ✅ | ✅ | Native plugin (recommended) |
-| **Copilot CLI** | ✅ | ✅ | ✅ | ✅ | Native plugin (recommended) |
-| **Cursor** | ✅ | — | — | — | install.sh (symlinks) |
-| **Cline** | ✅ | — | — | — | install.sh (symlinks) |
-| **Windsurf** | ✅ | — | — | — | install.sh (concatenated) |
-| **Gemini CLI** | ✅ | — | — | — | install.sh (symlink → AGENTS.md) |
-| **Claude.ai Web / Skills API** | — | ✅ | — | — | `task build-cloud-bundles-all` → `dist/cloud/<skill>.zip` |
-
-Skills use a `SKILL.md` format with YAML frontmatter, compatible with the
-[Agent Skills](https://agentskills.io) community spec and with Claude Code's
-Agent Skills specification. Cloud bundles produced by
-`scripts/build_cloud_bundle.py` follow the same format with cloud-side
-adjustments (description budget, sandbox note, package-internal path-swap).
-
----
-
-## Path resolution and Copilot integration
-
-Cross-references inside `.augment/rules/`, `.augment/skills/`, and the
-mirrored `.claude/`, `.cursor/`, `.clinerules/` trees use **relative
-paths from the delivered location**. They resolve through the symlinks
-created by `scripts/install.sh`, not via raw git checkout. This means
-GitHub Copilot's static checker — which walks the git tree — will see
-broken paths where there are none. **The gap is intentional, not a bug.**
-
-The package ships **one** Copilot instruction artefact:
-
-| File | Read by | Purpose |
-|---|---|---|
-| `.github/copilot-instructions.md` | Copilot Chat + PR review | Repo-wide coding standards plus the path-resolution suppression floor (Known False Positives) |
-
-Per [GitHub's documented convention](https://docs.github.com/en/copilot/reference/custom-instructions-support),
-Copilot Code Review reads `.github/copilot-instructions.md`
-repository-wide and `.github/instructions/**/*.instructions.md` for
-path-specific rules — there is no separate "review-only" instruction
-file. The first 4000 characters are the budget; keep
-high-priority rules (Scope Control, Known False Positives) up top.
-
-Installed (copy-if-missing) by `scripts/install.sh` from
-`src/agent-src/templates/`. Consumers can edit it freely;
-the installer never overwrites.
-
-The mechanical floor is `scripts/check_condensed_paths.py`, wired into
-`task ci` as `check-condensed-paths`. It validates `dist/agent-src/rules/*.md`:
-
-- `load_context:` entries must resolve to existing files.
-- Forbidden substrings (`.agent-src.uncondensed/`, `../../docs/`,
-  `../../agents/`) must not survive condensation — unless declared
-  per-rule via the `validator_ignore:` frontmatter primitive (audited).
-- Body links to `../docs/guidelines/...` are intentionally **not**
-  checked (they are package-internal reference material, silenced by
-  the Copilot suppression floor above).
-
-### Verifying path fixes in a consumer
-
-If a regression is suspected, replay the smoke test against the
-package's own `.augment/` projection:
-
-```bash
-task sync                              # regenerate dist/agent-src/ → .augment/
-./scripts-run src/scripts/smoke_path_resolution
-```
-
-The script walks `.augment/rules/*.md` and resolves every
-`load_context:` entry to a file under `.augment/`. A non-zero exit
-means a consumer would also see the same broken reference.
-
----
-
-← [Back to README](../README.md)