Review rules/constraints of this project

Author: johnteee

CONTRIBUTING.md

@@ -10,10 +10,11 @@ TeaPrompt is a **reflective engineering prompt library** focused on quality over
 
 ## Harness Policy (Nine Skills)
 
-TeaPrompt ships **nine frozen workflow skills** as natural-language harness policy — not a multi-agent runtime.
+TeaPrompt ships **nine frozen core workflow skills** as natural-language harness policy — not a multi-agent runtime. Nine is the current bounded routing set, not an optimality claim; optional registered domain packs remain outside core routing.
 
-- Pick **Strictness L1–L6** before choosing a skill ([reflective-dispatch](reflective-prompt-library/skills/reflective-dispatch/SKILL.md)).
-- Do **not** add a tenth core skill without the three-recurrence promotion gate ([PROJECT_KNOWLEDGE](reflective-prompt-library/PROJECT_KNOWLEDGE.md)).
+- Pick **Strictness L1–L6** before choosing a core skill ([reflective-dispatch](reflective-prompt-library/skills/reflective-dispatch/SKILL.md)).
+- Do **not** add a tenth core skill without the three-recurrence promotion gate and explicit human approval ([PROJECT_KNOWLEDGE](reflective-prompt-library/PROJECT_KNOWLEDGE.md)).
+- Add or extend a domain pack only through the canonical admission rule in [`06-repo/AGENTS.md`](reflective-prompt-library/06-repo/AGENTS.md#harness-policy-nine-skills); keep it out of `reflective-dispatch` routes and update `DOMAIN_PACK_SKILLS`, its ledger, demotion triggers, and structural guards.
 - Routing fairness: [ROUTING_CONTRACT](reflective-prompt-library/plans/ROUTING_CONTRACT.md); panel record: [multi-agent-panel-consensus](reflective-prompt-library/plans/multi-agent-panel-consensus-2026-06-25.md).
 
 ```bash

README.md

@@ -14,7 +14,7 @@ Doing the right thing > doing things right.
 
 ## North Star
 
-TeaPrompt helps humans and host agents choose the **right amount of rigor** for a task, record **why** decisions were made, and verify outcomes with **evidence** — using composable prompt layers and nine workflow skills as natural-language harness policy, **without** operating its own agent runtime.
+TeaPrompt helps humans and host agents choose the **right amount of rigor** for a task, record **why** decisions were made, and verify outcomes with **evidence** — using composable prompt layers, [nine core workflow skills plus optional registered domain packs](reflective-prompt-library/skills/skill-map.md) as natural-language harness policy, **without** operating its own agent runtime.
 
 Full library docs: [reflective-prompt-library/README.md](reflective-prompt-library/README.md).
 

reflective-prompt-library/06-repo/AGENTS.md

@@ -6,8 +6,8 @@ Repository-level harness policy for reflective engineering agents. Primary workf
 
 ## Scope
 
-- In scope: strictness-first routing, nine frozen workflow skills (frozen means gated, not never), evidence-backed completion, project-knowledge authority boundary.
-- Out of scope: multi-agent runtime, tenth core skill without promotion gate, silent rigor downgrade.
+- In scope: strictness-first routing, nine frozen core workflow skills (the current bounded routing set; frozen means gated, not never), registered domain packs outside core routing, evidence-backed completion, project-knowledge authority boundary.
+- Out of scope: multi-agent runtime, tenth core skill without promotion gate, unregistered or silently core-routed domain packs, silent rigor downgrade.
 
 ## Acceptance Criteria
 
@@ -41,14 +41,26 @@ runtime. For non-trivial routing, governance, or skill-selection work:
 
 1. **Strictness before skills** — pick L1–L6 first, then the smallest workflow.
    See [README.md](../README.md#strictness-before-skills) and [GLOSSARY.md](../GLOSSARY.md).
-2. **Nine frozen workflow skills** — do not add a tenth core skill without the
-   three-recurrence promotion gate ([PROJECT_KNOWLEDGE.md](../PROJECT_KNOWLEDGE.md)).
-3. **Fast routing** — [skills/SKILL_TRIGGER_CHEATSHEET.md](../skills/SKILL_TRIGGER_CHEATSHEET.md)
+2. **Nine frozen core workflow skills** — nine is the current cardinality, not
+   an optimality claim; the load-bearing invariant is a bounded registered core
+   routing set. Do not add a tenth core skill without the three-recurrence
+   promotion gate and explicit human approval
+   ([PROJECT_KNOWLEDGE.md](../PROJECT_KNOWLEDGE.md)).
+3. **Registered domain packs (not core routing)** — only skills listed in
+   `plans/validate_skill_examples.py` `DOMAIN_PACK_SKILLS` may extend `skills/`
+   outside the core set. They are host-invoked contracts, not selectable by
+   `reflective-dispatch` route rows. Adding or extending one requires explicit
+   human approval, a panel/decision record with Candidate Adoption Ledger,
+   demotion triggers, and registry/guard updates. An explicit user instruction
+   may supply approval, but recurrence evidence stays `unknown` unless observed;
+   this exception never waives the tenth-core promotion gate. See
+   [skills/skill-map.md](../skills/skill-map.md#registered-domain-packs-not-core-routing).
+4. **Fast routing** — [skills/SKILL_TRIGGER_CHEATSHEET.md](../skills/SKILL_TRIGGER_CHEATSHEET.md)
    (繁中: [`.zh-TW.md`](../skills/SKILL_TRIGGER_CHEATSHEET.zh-TW.md)).
-4. **Fairness contract** — equivalent intent must route equivalently with visible
+5. **Fairness contract** — equivalent intent must route equivalently with visible
    route trace; see [plans/ROUTING_CONTRACT.md](../plans/ROUTING_CONTRACT.md).
-5. **Panel / multi-voice rethink** — optional method inside
-   `reflective-research` (single host, persona lenses); judgment record at
+6. **Panel / multi-voice rethink** — optional method inside
+   `reflective-research` (single host, role-based lenses); judgment record at
    [plans/multi-agent-panel-consensus-2026-06-25.md](../plans/multi-agent-panel-consensus-2026-06-25.md).
 
 Run `make all` from the repository root before claiming routing or governance changes are verified.

reflective-prompt-library/METHODOLOGY_MAP.md

@@ -69,7 +69,7 @@ current decision legible?"**
 
 ## The 10-Layer Taxonomy
 
-The taxonomy has ten layers for project surfaces; the execution routing set remains the nine frozen workflow skills under `skills/`. Skill-count decisions follow the nine-skill policy, not the layer count. Frozen means gated, not never: a tenth core workflow skill needs the three-recurrence promotion gate and explicit human approval.
+The taxonomy has ten layers for project surfaces; the execution routing set remains nine frozen **core** workflow skills under `skills/`. Nine is the current bounded-set cardinality, not an optimality claim. Frozen means gated, not never: a tenth core workflow skill needs the three-recurrence promotion gate and explicit human approval. Optional registered domain packs are host-invoked contracts outside this routing set; see [`skills/skill-map.md`](skills/skill-map.md#registered-domain-packs-not-core-routing).
 
 Instead of a single prompt, the methodology is formally structured into ten distinct layers:
 
@@ -131,9 +131,9 @@ are stable and a human confirms the promotion.
 ## Multi-Agent Panel Consensus Addendum
 
 The 2026-06-25 multi-lens Socratic panel (six provider perspectives, single host)
-reaffirmed the north star: **nine frozen workflow skills**, strictness-first routing,
-evidence-backed evals in CI, optional multi-voice synthesis in `reflective-research`
-— not a swarm runtime. Judgment record:
+reaffirmed the north star: a bounded set of **nine frozen core workflow skills**,
+strictness-first routing, evidence-backed evals in CI, and optional multi-voice
+synthesis in `reflective-research` — not a swarm runtime. Judgment record:
 [plans/multi-agent-panel-consensus-2026-06-25.md](plans/multi-agent-panel-consensus-2026-06-25.md).
 
 ## Strictness Levels
@@ -233,6 +233,6 @@ Multi-agent work requires state, tools, logs, verification, and handoff artifact
 | Workflow & Agentic Layer | Aligned | Extended | Workflows, SOP compiler planning, planning engine, and memory consolidation (`04-agent/`). |
 | Domain Pack Layer | Aligned | Complete | Keep strategy and domain prompts in `05-domain/`. |
 | Repository Template Layer | Aligned | Extended | Maintain instruction templates and the reusable project-knowledge scaffold under `06-repo/`. |
-| Skill / Action Layer | Aligned | Extended | Maintain nine frozen workflow skills (including reflective-minimality gate) for anti-bloat decisions. |
+| Skill / Action Layer | Aligned | Extended | Maintain nine frozen core workflow skills (including reflective-minimality gate) plus separately registered, host-invoked domain packs outside core routing. |
 | Quality Gate / Verification | Aligned | Complete | Standardize skill-level quality checks. |
 | Governance / Capability Risk | Aligned | Extended | Use `reflective-risk` for high-risk boundaries and `04-agent/runtime-trust-boundary.md` for instruction/data/tool authority review. |

reflective-prompt-library/PROJECT_KNOWLEDGE.md

@@ -34,8 +34,9 @@ Language: English
 
 TeaPrompt helps humans and host agents choose the right amount of rigor for a task,
 record why decisions were made, and verify outcomes with evidence — using composable
-prompt layers and nine workflow skills as natural-language harness policy, without
-operating its own agent runtime. See [README.md](README.md#north-star).
+prompt layers, nine core workflow skills, and optional registered domain packs as
+natural-language harness policy, without operating its own agent runtime. See
+[README.md](README.md#north-star).
 
 > The active problem the project is solving now, with explicit non-goals so scope
 > does not silently expand. Completed milestones are retired from this section;
@@ -49,9 +50,10 @@ deferred promotions are recurrence-gated — see [panel backlog](plans/multi-age
 
 ### Standing Non-Goals
 - TeaPrompt does not operate its own multi-agent runtime, swarm, async peer messaging layer, recorder, replay engine, or side-effect enforcer; runtime designs remain references unless project direction changes.
-- The core workflow layer remains nine frozen workflow skills; a tenth core skill needs the three-recurrence promotion gate and explicit human approval.
+- The core routing layer remains nine frozen workflow skills; nine is the current bounded-set cardinality, not an optimality claim. A tenth core skill needs the three-recurrence promotion gate and explicit human approval.
+- Registered domain packs are host-invoked contracts outside core routing and follow the binding admission rule in [06-repo/AGENTS.md](06-repo/AGENTS.md#harness-policy-nine-skills); a user-directed exception records recurrence as `unknown`, never as proof of recurrence.
 - Full `SKILL.md` localization is not current direction; zh-TW supports navigation, cheatsheet routing, and glossary lines while English stays canonical for skill contracts.
-- Seeded routing evals are regression guards, not proof of general semantic routing quality; holdout expansion precedes router tuning.
+- Router tuning follows [ROUTING_CONTRACT.md](plans/ROUTING_CONTRACT.md) R8: add holdout evidence before tuning. Seeded routing evals are regression guards, not proof of general semantic routing quality.
 - "Methodology-complete" does not mean "operationally complete"; prompt/spec coverage does not promise persistence, replay, cancellation, idempotency, or role isolation.
 - Skills and prompt lenses may specify required runtime guarantees as preconditions, but TeaPrompt does not enforce or warrant them; host-runtime code and tests are the authority for operational guarantees, and missing enforcement evidence is `unknown` / no-go.
 

reflective-prompt-library/README.md

@@ -21,13 +21,13 @@ Core identity prompt
 
 ## North Star
 
-TeaPrompt helps humans and host agents choose the **right amount of rigor** for a task, record **why** decisions were made, and verify outcomes with **evidence** — using composable prompt layers and nine workflow skills as natural-language harness policy, **without** operating its own agent runtime.
+TeaPrompt helps humans and host agents choose the **right amount of rigor** for a task, record **why** decisions were made, and verify outcomes with **evidence** — using composable prompt layers, [nine core workflow skills plus optional registered domain packs](skills/skill-map.md) as natural-language harness policy, **without** operating its own agent runtime.
 
 ## Strictness Before Skills
 
 Pick **Strictness L1–L6** first (`skills/reflective-dispatch/SKILL.md`, [GLOSSARY.md](GLOSSARY.md)), then the skill phase. Skills declare `context_load: low|medium|high` in frontmatter so cost-aware hosts can defer heavy skills when Strictness is low.
 
-Standing non-goals are explicit: no TeaPrompt-owned runtime/swarm, no tenth core workflow skill without promotion evidence (frozen means gated, not never), no full `SKILL.md` i18n by default, and no silent rigor downgrade. Seeded routing evals are regression guards, not semantic-routing proof; methodology-complete does not mean operationally complete; skills may declare required runtime guarantees as preconditions, but host-runtime code and tests must enforce them. See [PROJECT_KNOWLEDGE.md](PROJECT_KNOWLEDGE.md#standing-non-goals).
+Standing non-goals are explicit: no TeaPrompt-owned runtime/swarm, no tenth core workflow skill without promotion evidence (frozen means gated, not never), and no domain pack inside core routing; optional packs follow the admission rule in `06-repo/AGENTS.md`. Full `SKILL.md` i18n is not a default, and silent rigor downgrade is forbidden. Seeded routing evals are regression guards, not semantic-routing proof; methodology-complete does not mean operationally complete; skills may declare required runtime guarantees as preconditions, but host-runtime code and tests must enforce them. See [PROJECT_KNOWLEDGE.md](PROJECT_KNOWLEDGE.md#standing-non-goals).
 
 
 ## Governance Panel Record
@@ -44,9 +44,9 @@ Multi-agent Socratic consensus on project goals and the nine skills (Rounds 1–
 - `05-domain/`: high-risk review, research, business strategy, learning, writing, and creative template prompts.
 - `06-repo/`: repository-level instruction templates plus a reusable
   `PROJECT_KNOWLEDGE.template.md` for project-specific design judgement.
-- `skills/`: concise `SKILL.md` workflow wrappers that map the prompt library into practical agent workflows.
-- `skills/SKILL_TRIGGER_CHEATSHEET.md`: one-page trigger cues and anti-misuse rules for the 9 workflow skills.
-- `skills/examples/`: worked input/output examples for each workflow skill.
+- `skills/`: nine core `SKILL.md` workflow wrappers plus optional registered domain packs; [skill-map.md](skills/skill-map.md) is the tier map.
+- `skills/SKILL_TRIGGER_CHEATSHEET.md`: one-page trigger cues and anti-misuse rules for the 9 core workflow skills, with a host-invoked domain-pack appendix.
+- `skills/examples/`: worked input/output examples for every registered core or domain-pack skill.
 - `plans/`: decision records, research reflections, eval fixtures, and governance contracts; live pointers are summarized in [PROJECT_KNOWLEDGE.md](PROJECT_KNOWLEDGE.md).
 - `plans/ROUTING_CONTRACT.md`: routing fairness, observability, and risk-based default-up contract.
 - `plans/route-001-paraphrase-eval.yaml`: paraphrase routing eval seed for ROUTE-001.