@event4u/agent-config 6.1.0 → 7.0.0

package/docs/SKILL_CENSUS.md

@@ -0,0 +1,344 @@
+# Skill-Family Census
+
+> **Status: provisional.** First-pass family assignment; the Keep / Prune / Merge
+> calls are description-overlap heuristics, not decisions. This document is input
+> for the leanness track (`road-to-tier-removal` and the command-surface-leanness
+> track), not a directive. Family assignments are derived from each skill's `name`
+> + `description` frontmatter only — no skill body was read. Empty families after
+> pruning are a merge signal, not a failure (per the Phase-0 council convergence
+> in `road-to-contract-integrity`).
+
+Produced as Phase-0 step F2 of `road-to-contract-integrity`. Covers all 237
+skills under `src/skills/*/SKILL.md`.
+
+## Families
+
+Every skill is assigned to exactly one of the fixed family set:
+
+`engineering`, `review`, `security`, `product`, `finance`, `content`, `video`,
+`agent-admin`, `evidence`, `meta-config`.
+
+Family semantics used for this pass:
+
+- **engineering** — writing / editing / debugging / testing application code,
+  framework carve-outs, infra/devops, code-level analysis, perf, DB, CI.
+- **review** — reviewing or auditing *someone's change* (code, design, PR,
+  decision, README) and the judge/critique lenses.
+- **security** — threat modeling, authz, secrets, privacy, data-handling,
+  security audit (the application-security surface, not agent-config security).
+- **product** — product/PM, growth, GTM, customer/discovery, people/org,
+  positioning, prioritisation, planning that is not code.
+- **finance** — cash, valuation, forecasting, unit-economics, fundraising.
+- **content** — prose/marketing/brand writing, docs co-authoring, editorial,
+  prompt engineering for end-users, image-still generation.
+- **video** — AI video / animation pipeline (scene, motion, character, song).
+- **agent-admin** — authoring/maintaining the agent-config artifacts themselves
+  (skills, rules, commands, personas, memory, roadmaps, the toolchain).
+- **evidence** — codebase orientation / project-intelligence discovery surfaces
+  whose output is an evidence write-up rather than a code change.
+- **meta-config** — cross-cutting reasoning/orchestration/execution-discipline
+  skills, plus the catch-all for skills that fit no other family.
+
+## Per-family count summary
+
+| Family | Count | Prune | Merge |
+|---|---:|---:|---:|
+| engineering | 83 | 0 | 7 |
+| product | 37 | 0 | 2 |
+| agent-admin | 32 | 1 | 5 |
+| content | 20 | 0 | 2 |
+| review | 18 | 0 | 2 |
+| meta-config | 15 | 0 | 1 |
+| security | 10 | 0 | 1 |
+| evidence | 9 | 0 | 1 |
+| finance | 7 | 0 | 0 |
+| video | 6 | 0 | 0 |
+| **Total** | **237** | **1** | **21** |
+
+## Census table
+
+| Skill | Family | Keep / Prune / Merge | Note |
+|---|---|---|---|
+| adr-create | agent-admin | Keep | |
+| agent-docs-writing | agent-admin | Keep | |
+| agents-md-thin-root | agent-admin | Keep | |
+| ai-council | agent-admin | Keep | |
+| check-refs | agent-admin | Keep | |
+| command-routing | agent-admin | Keep | |
+| command-writing | agent-admin | Keep | |
+| condense-memory | agent-admin | Keep | |
+| context-authoring | agent-admin | Merge → context-document | Two skills for filling/structuring context files; consider folding. |
+| context-document | agent-admin | Merge ← context-authoring | Merge target for context-authoring. |
+| conventional-commits-writing | agent-admin | Keep | commit-message conventions; borders engineering (git surface). |
+| decision-record | agent-admin | Merge → adr-create | decision-record frames the trade-off; adr-create writes the ADR — overlap on ADR drafting. |
+| description-assist | agent-admin | Keep | |
+| emit-tickets | agent-admin | Keep | roadmap → ticket-bundle materialisation. |
+| guideline-writing | agent-admin | Keep | |
+| learning-to-rule-or-skill | agent-admin | Keep | |
+| lint-skills | agent-admin | Keep | |
+| md-language-check | agent-admin | Keep | |
+| memory-consolidation | agent-admin | Keep | |
+| override-management | agent-admin | Keep | |
+| persona-writing | agent-admin | Keep | |
+| project-docs | agent-admin | Keep | |
+| roadmap-management | agent-admin | Merge ← roadmap-writing | Authoring vs. lifecycle-management of roadmaps — heavy surface overlap. |
+| roadmap-writing | agent-admin | Merge → roadmap-management | Merge candidate with roadmap-management. |
+| rule-refactor | agent-admin | Keep | |
+| rule-writing | agent-admin | Keep | |
+| script-writing | agent-admin | Keep | |
+| skill-improvement-pipeline | agent-admin | Prune | Orchestrates capture/classify/create/validate/apply — thin wrapper over skill-management + learning-to-rule-or-skill. |
+| skill-management | agent-admin | Keep | |
+| skill-reviewer | agent-admin | Keep | overlaps with skill-management/skill-writing but is the review lens; keep for now. |
+| skill-writing | agent-admin | Keep | |
+| upstream-contribute | agent-admin | Keep | contribute a learning back to the shared package. |
+| canvas-design | content | Keep | static visual / poster design. |
+| content-funnel-design | content | Keep | |
+| design-intelligence | content | Keep | grounded design brief (style/tokens/typography); borders engineering. |
+| doc-coauthoring | content | Keep | |
+| editorial-calendar | content | Keep | |
+| image-analyser | content | Keep | character-image analysis; borders video but reusable for stills. |
+| image-creator | content | Keep | character-image generation to spec. |
+| markitdown | content | Keep | doc→markdown conversion for ingestion. |
+| messaging-architecture | content | Keep | message stack from positioning. |
+| prompt-engineering-patterns | content | Merge ← prompt-optimizer | production-LLM prompt patterns; overlaps the prompt-* cluster. |
+| prompt-optimizer | content | Merge → prompt-engineering-patterns | end-user prompt rewriting; prompt-* cluster overlap. |
+| readme-writing | content | Keep | |
+| readme-writing-package | content | Keep | package-specific variant of readme-writing; distinct enough to keep this pass. |
+| refine-prompt | content | Keep | engine-facing prompt reconstruction (distinct from end-user optimize). |
+| release-comms | content | Keep | |
+| repomix-packer | content | Keep | codebase→single-file packing for LLM ingestion. |
+| technical-specification | content | Keep | spec/PRD/RFC authoring. |
+| voc-extract | content | Keep | Voice-of-Customer theme extraction. |
+| voice-and-tone-design | content | Keep | brand voice. |
+| corpus-grounding | evidence | Keep | shared grounding engine; could also read as meta-config. |
+| project-analysis-core | evidence | Keep | discovery primitives. |
+| project-analysis-hypothesis-driven | evidence | Keep | |
+| project-analysis-laravel | evidence | Keep | framework-specific deep analysis. |
+| project-analysis-nextjs | evidence | Keep | |
+| project-analysis-node-express | evidence | Keep | |
+| project-analysis-react | evidence | Keep | |
+| project-analysis-symfony | evidence | Keep | |
+| project-analysis-zend-laminas | evidence | Merge → project-analysis-symfony | Legacy Zend/Laminas; low-traffic stack — fold into a generic PHP analysis pass if cull warrants. |
+| accessibility-auditor | engineering | Keep | a11y review of UI code. |
+| api-design | engineering | Keep | |
+| api-endpoint | engineering | Merge → laravel-api-endpoint | api-endpoint is a stack-router that mostly dispatches to laravel-api-endpoint/nextjs-patterns. |
+| api-testing | engineering | Keep | |
+| artisan-commands | engineering | Keep | |
+| async-python-patterns | engineering | Keep | |
+| aws-infrastructure | engineering | Keep | |
+| blade-ui | engineering | Keep | |
+| code-refactoring | engineering | Keep | |
+| complexity-first-planning | engineering | Keep | risk-first decomposition; borders meta-config. |
+| composer-packages | engineering | Keep | |
+| dashboard-design | engineering | Keep | monitoring-dashboard design; borders content (data storytelling). |
+| database | engineering | Keep | |
+| defense-in-depth | engineering | Keep | validation guard design; borders security. |
+| dependency-upgrade | engineering | Keep | |
+| design-tokens | engineering | Keep | DTCG token system; UI-engineering. |
+| developer-like-execution | engineering | Keep | |
+| docker | engineering | Keep | |
+| eloquent | engineering | Merge → laravel | Eloquent is a Laravel subsurface; could fold into laravel. |
+| error-handling-patterns | engineering | Keep | |
+| existing-ui-audit | engineering | Keep | UI inventory gate before UI edits. |
+| fe-design | engineering | Keep | frontend-design heuristics reference. |
+| feature-planning | engineering | Keep | idea→plan→roadmap for code work. |
+| finishing-a-development-branch | engineering | Keep | ship/merge/park routing. |
+| flux | engineering | Keep | |
+| form-handler | engineering | Keep | |
+| git-workflow | engineering | Keep | |
+| github-ci | engineering | Keep | |
+| grafana | engineering | Keep | dashboards/log queries; borders content/observability. |
+| jobs-events | engineering | Keep | |
+| laravel | engineering | Keep | |
+| laravel-api-endpoint | engineering | Keep | merge target for api-endpoint. |
+| laravel-dto | engineering | Keep | |
+| laravel-horizon | engineering | Keep | |
+| laravel-mail | engineering | Keep | |
+| laravel-middleware | engineering | Keep | |
+| laravel-migration | engineering | Keep | |
+| laravel-notifications | engineering | Keep | |
+| laravel-pennant | engineering | Keep | |
+| laravel-pulse | engineering | Keep | |
+| laravel-reverb | engineering | Merge → laravel-websocket | Reverb is the WebSocket server; overlaps laravel-websocket's broadcasting surface. |
+| laravel-scheduling | engineering | Keep | |
+| laravel-validation | engineering | Keep | |
+| laravel-websocket | engineering | Keep | merge target for laravel-reverb. |
+| livewire | engineering | Keep | |
+| livewire-architect | engineering | Keep | shaping vs. writing Livewire; distinct enough this pass. |
+| logging-monitoring | engineering | Merge → grafana | logging/monitoring overlaps grafana + sentry-integration; consolidate observability. |
+| merge-conflicts | engineering | Keep | |
+| migration-architect | engineering | Keep | rollout/cutover shaping; hands off DDL to laravel-migration. |
+| mobile-e2e-strategy | engineering | Keep | |
+| module-detect-on-the-fly | engineering | Keep | |
+| module-management | engineering | Keep | |
+| multi-tenancy | engineering | Keep | borders security (tenant isolation). |
+| nextjs-patterns | engineering | Keep | |
+| openapi | engineering | Keep | |
+| performance | engineering | Merge → performance-analysis | performance + performance-analysis cover the same optimization surface. |
+| performance-analysis | engineering | Merge ← performance | merge target. |
+| pest-testing | engineering | Keep | |
+| php-coder | engineering | Keep | |
+| php-debugging | engineering | Keep | |
+| php-service | engineering | Keep | |
+| playwright-architect | engineering | Keep | shaping vs. writing Playwright; distinct this pass. |
+| playwright-testing | engineering | Keep | |
+| project-analyzer | engineering | Keep | single-pass tech-stack detection; borders evidence. |
+| quality-tools | engineering | Keep | PHPStan / Rector / ECS output handling. |
+| react-native-setup | engineering | Keep | |
+| react-shadcn-ui | engineering | Keep | |
+| sentry-integration | engineering | Keep | error investigation; borders observability cluster. |
+| sql-writing | engineering | Keep | |
+| symfony-workflow | engineering | Keep | |
+| systematic-debugging | engineering | Keep | |
+| tailwind-engineer | engineering | Keep | |
+| tech-debt-tracker | engineering | Keep | |
+| terraform | engineering | Keep | |
+| terragrunt | engineering | Merge → terraform | terragrunt wraps terraform multi-env; candidate fold if cull warrants. |
+| test-driven-development | engineering | Keep | |
+| test-performance | engineering | Keep | |
+| testing-anti-patterns | engineering | Keep | |
+| traefik | engineering | Keep | |
+| ui-component-architect | engineering | Keep | |
+| universal-project-analysis | engineering | Keep | multi-pass audit orchestrator; borders evidence. |
+| using-git-worktrees | engineering | Keep | |
+| source-discovery | engineering | Keep | read-the-real-source evidence gate before coding. |
+| dcf-modeling | finance | Keep | |
+| forecast-accuracy | finance | Keep | sales-forecast call construction. |
+| forecasting | finance | Keep | finance-side forecast; distinct from forecast-accuracy (sales) this pass. |
+| fundraising-narrative | finance | Keep | |
+| runway-cognition | finance | Keep | |
+| scenario-modeling | finance | Keep | |
+| unit-economics-modeling | finance | Keep | |
+| activation-design | product | Keep | |
+| build-buy-partner | product | Keep | |
+| churn-prevention | product | Keep | |
+| comp-banding | product | Keep | people/comp. |
+| competitive-moat-analysis | product | Keep | |
+| competitive-positioning | product | Keep | NOTE: this is package-vs-peer comparison; borders agent-admin but reads as product strategy. |
+| contracts-cognition | product | Keep | legal-contract reading for risk/constraint; fits no listed family cleanly — parked in product as a business-judgment lens. |
+| customer-research | product | Keep | |
+| deal-qualification-meddic | product | Keep | sales. |
+| discovery-interview | product | Merge → customer-research | discovery-interview + customer-research both shape discovery interviews. |
+| expansion-playbook | product | Keep | |
+| funnel-analysis | product | Keep | |
+| gtm-launch | product | Keep | |
+| hiring-loop-design | product | Keep | people/org. |
+| market-entry-analysis | product | Keep | |
+| okr-tree-modeling | product | Keep | |
+| onboarding-design | product | Keep | customer onboarding. |
+| onboarding-program | product | Keep | employee onboarding; distinct audience from onboarding-design. |
+| one-on-one-cadence | product | Keep | people/org. |
+| org-design | product | Keep | |
+| perf-feedback-craft | product | Keep | people/org. |
+| pipeline-strategy | product | Keep | sales. |
+| po-discovery | product | Merge → customer-research | ticket-shaping/problem-framing; overlaps discovery cluster but PM-facing — soft merge candidate. |
+| positioning-strategy | product | Keep | |
+| premortem | product | Keep | pre-commit failure imagination; borders review/meta-config. |
+| rice-prioritization | product | Keep | |
+| retention-loops | product | Keep | |
+| stakeholder-tradeoff | product | Keep | |
+| throughput-vs-morale-tradeoff | product | Keep | people/org. |
+| vision-articulation | product | Keep | |
+| estimate-ticket | product | Keep | ticket sizing; sibling of refine-ticket. |
+| refine-ticket | product | Keep | ticket refinement before planning. |
+| validate-feature-fit | product | Keep | feature-fit check; borders review. |
+| blameless-post-mortem | product | Keep | incident facilitation; people-process. |
+| incident-commander | product | Keep | incident framing/comms. |
+| launch-readiness | product | Keep | pre-merge release checklist; borders engineering. |
+| jira-integration | product | Keep | ticket tooling. |
+| adversarial-review | review | Keep | |
+| architecture-review-lens | review | Keep | judge dispatched by /review-changes. |
+| bug-analyzer | review | Keep | root-cause analysis of reported bugs. |
+| code-review | review | Keep | |
+| decision-review | review | Keep | backward audit of a past decision. |
+| design-review | review | Keep | UI/UX review. |
+| judge-bug-hunter | review | Keep | |
+| judge-code-quality | review | Keep | |
+| judge-security-auditor | review | Keep | security judge lens (paired with security family). |
+| judge-test-coverage | review | Keep | |
+| readme-reviewer | review | Keep | |
+| receiving-code-review | review | Keep | processing inbound review feedback. |
+| requesting-code-review | review | Keep | self-review + PR framing. |
+| review-routing | review | Keep | reviewer suggestion + risk flag. |
+| risk-officer | review | Keep | pre-commit risk surfacing. |
+| root-cause-frameworks | review | Merge → bug-analyzer | 5-whys/fishbone tracing; overlaps bug-analyzer + blameless-post-mortem on RCA. |
+| verify-completion-evidence | review | Keep | completion-evidence gate. |
+| verify-repair-loop | review | Merge ← verify-completion-evidence | iterate-to-green; pairs tightly with verify-completion-evidence (soft merge candidate). |
+| agent-security-review | security | Keep | adversarial review of agent CONFIG (distinct surface from app security). |
+| authz-review | security | Keep | |
+| blast-radius-analyzer | security | Keep | enumerates dependencies before editing shared code; pre-edit pre-action gate (pairs with data-flow-mapper / threat-modeling). |
+| data-flow-mapper | security | Keep | data-flow tracing before edits. |
+| data-handling-judgment | security | Keep | classification/retention/transfer. |
+| privacy-review | security | Keep | GDPR/CCPA/HIPAA. |
+| secrets-management | security | Keep | |
+| security | security | Merge → security-audit | generic "security best practices" overlaps the audit + judge-security surfaces. |
+| security-audit | security | Keep | merge target; explicit audit request. |
+| threat-modeling | security | Keep | |
+| analysis-autonomous-mode | meta-config | Keep | autonomous deep-investigation orchestration. |
+| analysis-skill-router | meta-config | Keep | routes to project-analysis-* skills. |
+| copilot-agents-optimization | meta-config | Merge → copilot-config | both tune AGENTS.md / copilot-instructions surfaces. |
+| copilot-config | meta-config | Keep | merge target. |
+| devcontainer | meta-config | Keep | dev-environment wiring (not app code; not Copilot). |
+| reasoning-orchestrator | meta-config | Keep | reasoning-chain coordination. |
+| sequential-thinking | meta-config | Keep | structured decomposition. |
+| subagent-orchestration | meta-config | Keep | implementer/judge subagent modes. |
+| token-optimizer | meta-config | Keep | pre-action token-saving decision tree. |
+| rtk-output-filtering | meta-config | Keep | rtk CLI output filtering. |
+| file-editor | meta-config | Keep | opens edited files in the IDE. |
+| prompt-validator | meta-config | Keep | pre-spend AI-video prompt contradiction gate; borders video. |
+| character-consistency | video | Keep | |
+| mcp | meta-config | Keep | MCP server usage reference. |
+| mcp-builder | meta-config | Keep | building MCP servers; borders engineering. |
+| motion-choreographer | video | Keep | |
+| pixar-storyteller | video | Keep | |
+| scene-expander | video | Keep | |
+| song-to-script | video | Keep | |
+| video-director | video | Keep | |
+| deep-reading-analyst | content | Keep | long-form article analysis; borders product/review. |
+| prediction-pool-optimizer | meta-config | Keep | NOTE: niche domain skill (sports-pool tip optimization) — fits no listed family cleanly; parked in meta-config as the catch-all per the brief. Prune candidate for the leanness track (off-mission for an agent-config suite). |
+
+> **Catch-all note.** `prediction-pool-optimizer` is the one skill that fits no
+> family on name+description (it optimizes sports-prediction-pool tips). Per the
+> brief it is parked in `meta-config`; the leanness track should treat it as a
+> prune/extract candidate (likely off-mission for a coding-agent suite).
+
+## Token-budget reading
+
+The 200k-token context budget (Karpathy peer-review `needs-verification` item:
+*does a per-persona load approach practical token limits?*) is measured here
+against the **engineering** family as the representative persona-scoped load — it
+is the largest family (83 skills) and the most likely single-persona load for a
+backend/full-stack engineer.
+
+Method: sum the byte size of all 83 engineering-family `SKILL.md` files (as
+assigned in the census table), then approximate tokens as `chars / 4`.
+
+- engineering-family SKILL.md files: **83**
+- total bytes: **568,799** (~556 KB)
+- approximate tokens: **≈ 142,200** (`568,799 / 4`)
+- share of the 200k budget: **≈ 70 %**
+
+**Reading.** Loading *only* the engineering family's skill bodies consumes roughly
+70 % of the entire 200k context budget — before any rules, AGENTS.md, conversation
+history, tool definitions, or the file under work are added. A naive "load the
+whole persona's skills" strategy is therefore **not** economically viable at the
+current artifact count: the engineering persona alone would leave under ~60k tokens
+for everything else, and a session that touches two families (e.g. engineering +
+review) would blow the budget outright.
+
+This confirms the peer-review concern: **artifact count is coupled to runtime
+cost**, and F2/F5 are not hygiene — they govern whether persona-scoped context
+loads stay viable. Two implications for the leanness and presentation tracks:
+
+1. Per-persona loading must stay **lazy / on-demand** (load a skill body only when
+   its trigger fires), not eager — the family grouping is a *navigation* aid, not a
+   bulk-load unit.
+2. The engineering family (83 of 237 skills) is the most over-represented and the
+   highest-value pruning/merge target; the Merge candidates flagged above (the
+   `laravel-*` carve-outs, the `performance`/`performance-analysis` pair, the
+   observability cluster, `api-endpoint` → `laravel-api-endpoint`) are where the
+   cull should start.
+
+> Figures are approximate (`chars/4` token estimate, frontmatter-only family
+> assignment) and provisional; they establish the cost baseline that Phase 2's
+> presentation work and the leanness cull share.

package/docs/architecture/augment-projection.md

@@ -34,7 +34,7 @@ After projection, those paths resolve through the symlinks in
 | Project self-projection | `task sync` (`project_to_augment()` step) — [`taskfiles/content.yml:4`](../../taskfiles/content.yml) |
 | Standalone project-augment | `task project-augment` — [`taskfiles/content.yml:23`](../../taskfiles/content.yml) |
 | Consumer install | `scripts/install.sh` (delegates to `scripts/install.py`) |
-| Direct script | `python3 scripts/condense.py --project-augment` |
+| Direct script | `./scripts-run src/scripts/condense --project-augment` |
 
 ## Invariants
 

package/docs/architecture/multi-tool-projection.md

@@ -37,7 +37,7 @@ removes `.cursor/` on next `task generate-tools`.
 |---|---|
 | Regenerate all enabled | `task generate-tools` ([`taskfiles/content.yml:63`](../../taskfiles/content.yml)) |
 | Clean output | `task clean-tools` ([`taskfiles/content.yml:69`](../../taskfiles/content.yml)) |
-| Direct script | `python3 scripts/condense.py --generate-tools` |
+| Direct script | `./scripts-run src/scripts/condense --generate-tools` |
 | Consumer install | `scripts/install.sh` (calls `--generate-tools` after `--project-augment`) |
 
 ## Invariants
@@ -107,8 +107,8 @@ with every tool ID temporarily enabled in `agents/.agent-tools.yml`.
 Re-run the measurement after every change to the projection logic:
 
 ```bash
-python3 scripts/measure_projection_bytes.py --regenerate
-python3 scripts/measure_projection_bytes.py --json    # CI-friendly
+./scripts-run src/scripts/measure_projection_bytes --regenerate
+./scripts-run src/scripts/measure_projection_bytes --json    # CI-friendly
 ```
 
 ## Proving the pipeline

package/docs/architecture.md

@@ -146,15 +146,46 @@ note, package-internal path-swap, description budget, and the
 
 | Layer | Count | Purpose |
 |---|---|---|
-| **Skills** | 230 | On-demand expertise — stack analysis (Laravel · Symfony · Zend / Laminas · Next.js · React · Node), testing, Docker, API design, security, observability, … |
-| **Rules** | 83 | Always-active constraints — coding standards, scope control, verification, language-and-tone, agent-authority |
-| **Commands** | 147 | Slash-command workflows — `/commit`, `/create-pr`, `/fix ci`, `/optimize skills`, `/feature plan`, `/work`, `/implement-ticket`, `/condense`, … |
-| **Guidelines** | 81 | Reference material cited by skills — PHP patterns, Eloquent, Playwright, agent-infra, … |
+| **Skills** | 251 | On-demand expertise — stack analysis (Laravel · Symfony · Zend / Laminas · Next.js · React · Node), testing, Docker, API design, security, observability, … |
+| **Rules** | 88 | Always-active constraints — coding standards, scope control, verification, language-and-tone, agent-authority |
+| **Commands** | 161 | Slash-command workflows — `/commit`, `/create-pr`, `/fix ci`, `/optimize skills`, `/feature plan`, `/work`, `/implement-ticket`, `/condense`, … |
+| **Guidelines** | 82 | 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).
@@ -206,7 +237,7 @@ execution:
 Invoke a runtime-capable skill end-to-end:
 
 ```bash
-python3 scripts/runtime_dispatcher.py run --skill lint-skills
+./scripts-run src/scripts/runtime_dispatcher run --skill lint-skills
 ```
 
 A typed `ExecutionResult` (exit code, stdout, stderr, duration,
@@ -347,7 +378,7 @@ package's own `.augment/` projection:
 
 ```bash
 task sync                              # regenerate dist/agent-src/ → .augment/
-python3 scripts/smoke_path_resolution.py
+./scripts-run src/scripts/smoke_path_resolution
 ```
 
 The script walks `.augment/rules/*.md` and resolves every

package/docs/benchmark.md

@@ -4,74 +4,71 @@
 
 ## Honesty labels (read first)
 
-> 1. **Wrapper-lift on a fixed host (`claude-sonnet-4-6`), NOT model-vs-model.** Measures what the agent-config package does to ONE host model on a neutral fixture — not a capability ranking.
+> 1. **Wrapper-lift on a fixed host (`claude-haiku-4-5`), NOT model-vs-model.** Measures what the agent-config package does to ONE host model on a neutral fixture — not a capability ranking.
 > 2. **Discipline axis, not capability.** The headline is the *discipline* delta (did it stay minimal / verify / ask / not destroy / update downstream), not whether the goal was achievable.
-> 3. **PILOT — low statistical power (N=5 tasks × 1 seed(s)).** Directional only.
+> 3. **PILOT — low statistical power (N=2 tasks × 12 seed(s)).** Directional only.
 > 4. **Paired design**, errored runs excluded; McNemar (capability) + Wilcoxon signed-rank (discipline) + effect sizes.
 > 5. **Not comparable to SWE-bench / GAIA / Fable scores** — a different question entirely.
 
-## Gate verdict: **FALSIFIED-OR-INCONCLUSIVE**
+## Gate verdict: **PASS**
 
 - capability lift significant: `False`
-- discipline lift significant: `False`
+- discipline lift significant: `True`
 - status-bucket better (package vs vanilla): `False`
 
-> **Honest null at this scale.** On this micro-fixture pilot the bare host is *already* disciplined (vanilla discipline ≈ 1.0), so there is no headroom for the package to lift. Per the 2026-06-14 council this is NOT a full falsification — a complete gate requires a **complexity-stratified** run (micro / meso / multi-file fixtures) to see whether headroom appears at realistic scale. That run (meso/multi fixtures + ~17M-token budget) is the deferred follow-up. No lift is claimed.
+> **Measurable discipline lift (significant).** On the scope-creep / downstream-changes family, a weak host (`claude-haiku-4-5`) leaves the downstream caller un-updated / scope-creeps a large fraction of the time; the package reliably corrects it. The lift is significant on the discipline axis (Wilcoxon p<0.05, every discordant pair favouring the package) AND beats an **equal-length inert-prose placebo** — so it is the package's *content* (its `downstream-changes`/`scope-control` rules), NOT mere prompt-length, that helps. **Honest scope (empirically bounded):** the lift is **weak-host-specific** — a CLEAN strong-host run (`claude-sonnet-4-6`, same tasks, 8 seeds) scored vanilla = package = placebo = 1.00 (no headroom, package redundant). So the package helps a WEAK model that lacks the discipline; a strong model already has it. This matches the package's design thesis (strong hosts self-apply discipline; weak hosts benefit fully). Discipline axis, not capability (both arms make the primary change); this task family (scope/downstream), not a universal claim. It improves *solution discipline*, not model intelligence.
 
-## package lift — `package` vs `vanilla` (n=5 pairs)
+## package lift — `package` vs `vanilla` (n=24 pairs)
 
 ### Table 1 — capability axis (expected near-flat by design)
 
 | metric | baseline | treatment | test |
 |---|---|---|---|
-| pass-rate | 80% | 80% | McNemar p=1.0, h=0.0 |
+| pass-rate | 100% | 100% | McNemar p=1.0, h=0.0 |
 
 ### Table 2 — discipline axis (the lift)
 
 | metric | baseline | treatment | Δ | test |
 |---|---|---|---|---|
-| mean discipline | 1.000 | 0.800 | -0.200 | Wilcoxon p=1.0, rb=-1.0 (n≠0=1) |
+| mean discipline | 0.333 | 1.000 | +0.667 | Wilcoxon p=0.0005, rb=1.0 (n≠0=16) |
 
-## RDP lift — `package-rdp` vs `package` (n=5 pairs)
+### Table 3 — cost axis (mean tokens/run, non-errored)
 
-### Table 1 — capability axis (expected near-flat by design)
-
-| metric | baseline | treatment | test |
+| metric | baseline | treatment | Δ |
 |---|---|---|---|
-| pass-rate | 80% | 80% | McNemar p=1.0, h=0.0 |
+| mean tokens | 90,534 | 992,044 | +901,510 |
 
-### Table 2 — discipline axis (the lift)
-
-| metric | baseline | treatment | Δ | test |
-|---|---|---|---|---|
-| mean discipline | 0.800 | 0.800 | +0.000 | Wilcoxon p=1.0, rb=0.0 (n≠0=0) |
-
-## attribution (content vs length) — `package` vs `placebo` (n=5 pairs)
+## attribution (content vs length) — `package` vs `placebo` (n=24 pairs)
 
 ### Table 1 — capability axis (expected near-flat by design)
 
 | metric | baseline | treatment | test |
 |---|---|---|---|
-| pass-rate | 60% | 80% | McNemar p=1.0, h=0.4421 |
+| pass-rate | 100% | 100% | McNemar p=1.0, h=0.0 |
 
 ### Table 2 — discipline axis (the lift)
 
 | metric | baseline | treatment | Δ | test |
 |---|---|---|---|---|
-| mean discipline | 0.900 | 0.800 | -0.100 | Wilcoxon p=1.0, rb=-0.3333 (n≠0=2) |
+| mean discipline | 0.333 | 1.000 | +0.667 | Wilcoxon p=0.0005, rb=1.0 (n≠0=16) |
+
+### Table 3 — cost axis (mean tokens/run, non-errored)
+
+| metric | baseline | treatment | Δ |
+|---|---|---|---|
+| mean tokens | 97,528 | 992,044 | +894,516 |
 
 ## Status buckets (trajectory)
 
 | arm | runs | error-rate | buckets |
 |---|---|---|---|
-| vanilla | 5 | 0% | completed:5 |
-| package | 5 | 0% | completed:5 |
-| package-rdp | 5 | 0% | completed:5 |
-| placebo | 5 | 0% | completed:5 |
+| vanilla | 24 | 0% | completed:24 |
+| package | 24 | 0% | completed:24 |
+| placebo | 24 | 0% | completed:24 |
 
 ## Methodology
 
-- Host model: `claude-sonnet-4-6` (pinned across all arms — a validity requirement, not a model comparison).
+- Host model: `claude-haiku-4-5` (pinned across all arms — a validity requirement, not a model comparison).
 - Per-run budget cap: $3.5; placebo injected ~6628 chars of inert prose.
 - Arms: vanilla (plugin off) · package (real plugin) · package-rdp (plugin + RDP rules) · placebo (plugin off + equal-length inert prose).
 - Corpus: `internal/bench/corpora/ab-trackb-v2.yaml` (5 trap archetypes). Scoring: `bench_ab_scoring_v2.py` (deterministic, no LLM judge).

package/docs/capability-matrix.md

@@ -0,0 +1,32 @@
+# Capability matrix — what works on which host
+
+> **Generated** by `scripts/generate_capability_matrix.py` — do NOT
+> hand-edit. Derived from the `generate_tools()` projection logic in
+> `condense.py` (each cell traces to a `generate_*` dispatcher call).
+> Drift-checked in CI (`--check`).
+
+Cells: **✅ native** (host consumes the artifact directly — symlink /
+native dir) · **🔁 adapter** (projected through a host-specific
+transform — `.mdc`, workflow, or an aggregated single file) · **— none**
+(no generator emits this artifact for this host).
+
+| Artifact | claude-code | claude-plugin | augment | cursor | windsurf | cline | gemini | copilot | claude-desktop |
+|---|---|---|---|---|---|---|---|---|---|
+| `rules` | ✅ native | — none | ✅ native | 🔁 adapter | 🔁 adapter | ✅ native | 🔁 adapter | 🔁 adapter † | — none |
+| `skills` | ✅ native | ✅ native | ✅ native | — none | — none | — none | — none | — none | — none |
+| `commands` | ✅ native | — none | ✅ native | 🔁 adapter | 🔁 adapter | — none | — none | — none | — none |
+| `personas` | ✅ native | — none | ✅ native | ✅ native | — none | — none | — none | — none | — none |
+| `user-types` | ✅ native | — none | ✅ native | ✅ native | — none | — none | — none | — none | — none |
+| `hooks` | — none | ✅ native | — none | — none | — none | — none | — none | — none | — none |
+
+## How to read this
+
+- Projection is **intentionally asymmetric** — a `— none` cell is a
+  design choice, not a bug. Skills project natively only where a host
+  has a native skill surface; everywhere else the rules + commands
+  carry the behaviour.
+- `🔁 adapter` cells are real coverage through a host-native shape
+  (Cursor `.mdc`, Windsurf workflows, the aggregated `GEMINI.md`).
+- `†` marks an **install-time** surface the installer writes (e.g.
+  `.github/copilot-instructions.md`), not the `generate_tools()` path —
+  real coverage, different code path.