@chrono-meta/fh-gate 1.4.41 → 1.4.42

package/knowledge/shared/harness-core/tpa_schema.md

@@ -0,0 +1,136 @@
+---
+name: tpa-schema
+description: Canonical Target Profile Analysis schema for FH meta-harness skills. All skills that run TPA (sim-conductor, steel-quench, phantom-quench, agent-composer) MUST derive their routing decisions from this schema.
+type: reference
+scope: meta-harness
+---
+
+# Target Profile Analysis (TPA) — Canonical Schema
+
+> **Meta-harness scope**: FH operates on harness assets themselves (SKILL.md, CLAUDE.md, agents, templates), not on application code. A mistake in a SKILL.md propagates to every future session that loads it — the risk profile is categorically higher than application code of the same line count.
+
+All four skills that perform TPA **reference this file** rather than defining their own taxonomy:
+- `sim-conductor` Step 0.3
+- `steel-quench` Step 0.3
+- `phantom-quench` Step 0.5
+- `agent-composer` Step 0.2
+
+---
+
+## Canonical Fields
+
+| Field | Type | Values | Derivation |
+|---|---|---|---|
+| `artifact_type` | enum | See §Artifact Type Table | From file path / extension / user declaration |
+| `audience` | enum | `internal` · `external` · `mixed` | From scope + external_publish |
+| `claim_density` | enum | `low` (≤3) · `medium` (4–10) · `high` (>10) | Count verifiable claims in artifact |
+| `risk_level` | enum | `low` · `medium` · `high` | From §Risk Default Table (override allowed) |
+| `novelty` | bool | `true` · `false` | `true` if no prior session evidence or first-of-its-kind pattern |
+| `scope` | enum | `internal` · `external` | `external` if published/publishable or public repo |
+| `phantom_risk` | derived | `true` · `false` | `true` if artifact contains arXiv/DOI/http citations OR code fences in a knowledge doc |
+
+> `phantom_risk` is **derived** — do not ask the user. If `artifact_type` is `skill_md` or `design_doc` and citations or URLs are present → `true`. Use as input to `phantom-quench` gate trigger.
+
+---
+
+## §Artifact Type Table (Meta-Harness First)
+
+### Meta-Harness Artifact Types
+
+| `artifact_type` | Examples | Default Risk | Why High |
+|---|---|:---:|---|
+| `skill_md` | `plugins/*/skills/*/SKILL.md` | **high** | Behavioral rule for AI — executes every session it's loaded |
+| `claude_md` | `CLAUDE.md`, `.claude/rules/*.md` | **high** | Governance layer — one drift affects all connected projects |
+| `agent_md` | `.claude/agents/*.md` | **medium-high** | Sub-agent behavioral rule — dispatched autonomously |
+| `template` | `templates/**` | **high** | Propagates to target projects — a mistake multiplies |
+| `skill_detail_md` | `*/SKILL_detail.md` | **medium** | On-demand reference — lower broadcast risk than SKILL.md |
+| `knowledge_doc` | `knowledge/shared/*.md` | **medium** | Reference only, but substantive carve-out applies |
+| `cheatsheet` | `CHEATSHEET.md` | **medium** | User-facing guide — external audience |
+| `readme` | `README.md` | **medium-high** | External facing — first impression for all users |
+| `bash_script` | `templates/*.sh`, `scripts/*.sh` | **high** | Executes in user environment |
+| `session_card` | `tracks/_meta/reference_next_session_starter.md` | **low** | Ephemeral state — low blast radius |
+| `catalog_entry` | `CATALOG.md` additions | **low** | Index only — no behavioral impact |
+
+### Standard Artifact Types (Field Project Work)
+
+| `artifact_type` | Examples | Default Risk |
+|---|---|:---:|
+| `code` | `.py`, `.ts`, `.go`, `.java` | **medium** |
+| `pr_diff` | git diff output | **medium** |
+| `config` | `.yaml`, `.json`, `.toml` | **medium-high** |
+| `api_spec` | OpenAPI, Proto | **medium** |
+| `test_suite` | `*_test.py`, `*.spec.ts` | **low** |
+| `design_doc` | ADRs, RFCs, architecture docs | **medium** |
+
+---
+
+## §Gate Routing Table
+
+Which gates are **mandatory** vs optional for each (artifact_type × risk_level) combination.
+
+| `artifact_type` | `scope` | steel-quench | phantom-quench | sim-conductor Area |
+|---|---|:---:|:---:|:---:|
+| `skill_md` | any | **MANDATORY** | **MANDATORY** if `phantom_risk=true` | D-skill |
+| `claude_md` / `agent_md` | any | **MANDATORY** | optional | D-skill (for agent_md) |
+| `template` | any | **MANDATORY** | **MANDATORY** | A (external publish check) |
+| `readme` / `cheatsheet` | external | **MANDATORY** | optional | **A MANDATORY** |
+| `bash_script` | any | **MANDATORY** (Wave 1 real-code first) | optional | — |
+| `knowledge_doc` | any | optional | **MANDATORY** if `phantom_risk=true` | — |
+| `code` / `pr_diff` | any | optional | optional | D-code |
+| `config` | external | **MANDATORY** | optional | — |
+| `design_doc` | external | **MANDATORY** | **MANDATORY** if citations | A |
+
+**Enforcement note**: "mandatory" means the gate BLOCKS the downstream step if skipped — not advisory. `pipeline-conductor` Step 0.5 implements this: HIGH severity OPEN chains from `return-path-gate` = FAIL (halt sweep), not CONDITIONAL_PASS. User override available via `S` but records as `degraded` and disqualifies `CLEAN (--full)`.
+
+---
+
+## §Profile Output Format
+
+All TPA-performing skills output this block (copy verbatim):
+
+```
+TPA Profile:
+  artifact_type: [type]
+  audience: [internal | external | mixed]
+  claim_density: [low | medium | high] ([N] claims)
+  risk_level: [low | medium | high]
+  novelty: [true | false]
+  scope: [internal | external]
+  phantom_risk: [true | false]
+  required_gates: [list from §Gate Routing Table]
+  optional_gates: [list]
+```
+
+Skills MAY add skill-specific derived fields (e.g., `steel-quench` adds `wave_weights`, `sim-conductor` adds `recommended_area`) but the canonical 8 fields above are the shared contract.
+
+---
+
+## Override Rules
+
+1. User may explicitly override `risk_level` — accepted without question.
+2. `scope` override: if artifact is in `templates/` or public repo, `scope=external` is forced regardless of user declaration.
+3. `phantom_risk` is derived-only — cannot be overridden to `false` if citations or code fences exist.
+4. `required_gates` cannot be overridden to empty — skip must be explicitly declared as `degraded: [gate] (user skipped)` in output.
+
+---
+
+## Meta-Harness Amplification Rule
+
+> A defect in a `skill_md` or `claude_md` artifact runs in every session that loads it — potentially hundreds of times before being caught. Apply the following multiplier when assessing severity:
+>
+> - `skill_md` / `claude_md` / `template`: **severity × 3** (broadcast multiplier)
+> - `agent_md`: **severity × 2**
+> - Standard artifact types: no multiplier
+
+This is why meta-harness defaults to `risk_level=high` for these types: the blast radius is not the artifact itself but every downstream invocation.
+
+---
+
+## Skill Integration Pointers
+
+| Skill | How to use this schema |
+|---|---|
+| `sim-conductor` Step 0.3 | Replace inline table with `artifact_type` lookup from §Artifact Type Table → `required_gates` from §Gate Routing Table |
+| `steel-quench` Step 0.3 | Derive `phantom_risk` per §Canonical Fields. Wave 5 activation condition: `scope=external AND risk_level=high` |
+| `phantom-quench` Step 0.5 | Gate trigger: `phantom_risk=true` OR `risk_level=high` |
+| `agent-composer` Step 0.2 | Capability fit scoring input: `artifact_type` determines `role_match` lookup domain |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chrono-meta/fh-gate",
-  "version": "1.4.41",
+  "version": "1.4.42",
   "description": "FH runtime adapters — run FH governance, skills, and agents via Claude or Codex with machine-parseable gates.",
   "license": "MIT",
   "keywords": [
@@ -61,7 +61,8 @@
     "plugins/fh-meta/agents",
     "plugins/fh-commons/skills",
     "plugins/fh-commons/agents",
-    "knowledge/shared/harness-core/fh_integration_contract.md",
+    "knowledge/shared/harness-core",
+    "knowledge/shared/dialogue",
     "README.md"
   ]
 }