@event4u/agent-config 2.25.0 → 3.0.0

package/scripts/prepack-check.mjs

@@ -0,0 +1,62 @@
+#!/usr/bin/env node
+/**
+ * prepack-check.mjs — guard the published artifact.
+ *
+ * Runs during `npm pack` and `npm publish`. Asserts the compiled TS
+ * CLI binary exists, is executable, and carries the Node shebang
+ * before the tarball is built. Otherwise a silently-broken `dist/`
+ * ships and every `npx @event4u/agent-config` greets the consumer
+ * with a cryptic `Cannot find module` panic.
+ *
+ * Folded from external council pass 2026-05-18 (Phase 1.3 acceptance).
+ *
+ * Skip with PREPACK_SKIP_BUILD_CHECK=1 only for local `npm pack` dry-runs
+ * that intentionally test the failure mode.
+ */
+import { readFileSync, statSync } from 'node:fs';
+import { resolve } from 'node:path';
+import { argv, env, exit } from 'node:process';
+
+const BIN = resolve('dist/cli/agent-config.js');
+const SHEBANG = '#!/usr/bin/env node';
+
+function die(msg) {
+    process.stderr.write(`prepack-check: ${msg}\n`);
+    exit(1);
+}
+
+if (env.PREPACK_SKIP_BUILD_CHECK === '1') {
+    process.stderr.write('prepack-check: skipped (PREPACK_SKIP_BUILD_CHECK=1)\n');
+    exit(0);
+}
+
+let st;
+try {
+    st = statSync(BIN);
+} catch {
+    die(`compiled CLI binary missing at ${BIN}. Run \`npm run build\` before \`npm pack\`.`);
+}
+
+if (!st.isFile()) {
+    die(`${BIN} is not a regular file.`);
+}
+
+// Executable bit check (skip on Windows where mode bits are unreliable).
+if (process.platform !== 'win32') {
+    // 0o111 = any executable bit set
+    if ((st.mode & 0o111) === 0) {
+        die(`${BIN} is not executable. Build step must chmod +x the bin entry.`);
+    }
+}
+
+const head = readFileSync(BIN, 'utf8').slice(0, SHEBANG.length);
+if (head !== SHEBANG) {
+    die(`${BIN} missing Node shebang. Expected "${SHEBANG}" as first line.`);
+}
+
+process.stderr.write(`prepack-check: ${BIN} OK\n`);
+
+// Optional: invoked with --verbose dumps the size for tarball-budget bookkeeping.
+if (argv.includes('--verbose')) {
+    process.stderr.write(`prepack-check: size=${st.size} bytes\n`);
+}

package/scripts/probe_projection_fidelity.py

@@ -5,7 +5,7 @@ Reads tests/fixtures/projection_fidelity/fixtures.yml, walks the
 projected trees (.augment/, .claude/, .cursor/, .clinerules/,
 .windsurfrules, .windsurf/), and records pass/fail/partial per check.
 
-Output: agents/reports/projection-fidelity.json + stdout summary.
+Output: agents/runtime/reports/projection-fidelity.json + stdout summary.
 
 Pure stdlib (PyYAML reuse from scripts/_lib if installed; otherwise
 inline minimal YAML loader for the fixture's restricted shape).
@@ -172,7 +172,7 @@ def check_entry(entry: dict) -> dict:
 def main() -> int:
     ap = argparse.ArgumentParser()
     ap.add_argument("--fixture", default="tests/fixtures/projection_fidelity/fixtures.yml")
-    ap.add_argument("--report", default="agents/reports/projection-fidelity.json")
+    ap.add_argument("--report", default="agents/runtime/reports/projection-fidelity.json")
     args = ap.parse_args()
 
     fixture = yaml.safe_load((ROOT / args.fixture).read_text(encoding="utf-8"))

package/scripts/refine_ticket_detect.py

@@ -94,7 +94,7 @@ class RepoContext:
 
     Feeds the skill's Top-5 risks with project-specific vocabulary —
     recent branch names (naming-convention signal), recent commit
-    subjects (active modules signal), and on-disk `agents/contexts/`
+    subjects (active modules signal), and on-disk `agents/settings/contexts/`
     documents (domain-vocabulary signal).
 
     Empty when `repo_aware=False` — the skill still produces the same

package/scripts/schemas/command.schema.json

@@ -2,7 +2,7 @@
   "$schema": "http://json-schema.org/draft-07/schema#",
   "$id": "https://github.com/event4u-app/agent-config/scripts/schemas/command.schema.json",
   "title": "Command frontmatter",
-  "$comment": "Source: agents/docs/frontmatter-contract.md#commands. `description` is required at schema level because 66/67 commands declare it — the one exception is a known gap, not a supported shape.",
+  "$comment": "Source: agents/reference/docs/frontmatter-contract.md#commands. `description` is required at schema level because 66/67 commands declare it — the one exception is a known gap, not a supported shape.",
   "type": "object",
   "additionalProperties": false,
   "required": ["name", "description", "disable-model-invocation"],
@@ -67,6 +67,11 @@
       "pattern": "^[0-9]+\\.[0-9]+\\.[0-9]+$",
       "description": "Semver release in which this command became a shim (e.g. '1.15.0')."
     },
+    "framework": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9-]*$",
+      "description": "Framework carve-out marker. Set on commands that are 100% coupled to one stack (e.g. `framework: laravel` on update-form-request-messages). Generic, multi-stack commands MUST omit this key. Routed by .agent-src.uncompressed/rules/framework-neutrality-in-generic-skills.md (Tier 2)."
+    },
     "council_depth": {
       "type": "string",
       "enum": ["deep"],
@@ -112,6 +117,46 @@
           "description": "Optional per-command override of the global cooldown duration (e.g. '10m'). Defaults to settings."
         }
       }
+    },
+    "workspaces": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "pattern": "^[a-z][a-z0-9-]*$"},
+      "description": "ADR-013 workspace ids (additive discovery axis). Closed vocabulary enforced by scripts/lint_artefact_frontmatter.py against config/discovery/workspaces.yml."
+    },
+    "packs": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "pattern": "^[a-z][a-z0-9-]*$"},
+      "description": "ADR-013 pack ids (additive discovery axis). Closed vocabulary enforced by scripts/lint_artefact_frontmatter.py against config/discovery/packs.yml."
+    },
+    "lifecycle": {
+      "type": "string",
+      "enum": ["active", "deprecated", "experimental", "archived"],
+      "description": "ADR-013 lifecycle state."
+    },
+    "trust": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["level", "confidence", "human_review_required"],
+      "properties": {
+        "level": {"type": "string", "enum": ["core", "professional", "experimental", "advisory", "restricted"]},
+        "confidence": {"type": "string", "enum": ["high", "medium", "low"]},
+        "human_review_required": {"type": "boolean"}
+      },
+      "description": "ADR-013 trust block. Vocabulary mirrored by scripts/lint_artefact_frontmatter.py."
+    },
+    "install": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["default", "removable"],
+      "properties": {
+        "default": {"type": "boolean"},
+        "removable": {"type": "boolean"}
+      },
+      "description": "ADR-013 install hints (default-on at consumer install; user-removable)."
     }
   }
 }

package/scripts/schemas/persona.schema.json

@@ -2,7 +2,7 @@
   "$schema": "http://json-schema.org/draft-07/schema#",
   "$id": "https://github.com/event4u-app/agent-config/scripts/schemas/persona.schema.json",
   "title": "Persona frontmatter",
-  "$comment": "Source: agents/docs/frontmatter-contract.md#personas. `mode` enum mirrors the values declared in personas/*.md plus the additional `incident` value from templates/persona.md.",
+  "$comment": "Source: agents/reference/docs/frontmatter-contract.md#personas. `mode` enum mirrors the values declared in personas/*.md plus the additional `incident` value from templates/persona.md.",
   "type": "object",
   "additionalProperties": false,
   "required": ["id", "role", "description", "tier", "mode", "version", "source"],

package/scripts/schemas/rule.schema.json

@@ -2,7 +2,7 @@
   "$schema": "http://json-schema.org/draft-07/schema#",
   "$id": "https://github.com/event4u-app/agent-config/scripts/schemas/rule.schema.json",
   "title": "Rule frontmatter",
-  "$comment": "Source: agents/docs/frontmatter-contract.md#rules. `type` is authoritative; `alwaysApply` is a Cursor/Cline sidecar and accepted but not enforced.",
+  "$comment": "Source: agents/reference/docs/frontmatter-contract.md#rules. `type` is authoritative; `alwaysApply` is a Cursor/Cline sidecar and accepted but not enforced.",
   "type": "object",
   "additionalProperties": false,
   "required": ["type", "source", "description"],
@@ -29,8 +29,8 @@
       "type": "array",
       "items": {
         "type": "string",
-        "pattern": "^((\\.\\./)*contexts/|agents/contexts/|\\.agent-src/contexts/)[^\\s]+\\.md$",
-        "description": "Logical name (preferred — `contexts/<area>/<file>.md`) or project-local (`agents/contexts/<file>.md`). The `.agent-src.uncompressed/` prefix is rejected by the regex; the rewriter (`scripts/compress.py::_rewrite_paths`) resolves logical names at compress time. Rewritten relative forms (`../contexts/...`, `../../contexts/...`) are accepted so the linter passes on the compressed mirror in CI."
+        "pattern": "^((\\.\\./)*contexts/|agents/settings/contexts/|\\.agent-src/contexts/)[^\\s]+\\.md$",
+        "description": "Logical name (preferred — `contexts/<area>/<file>.md`) or project-local (`agents/settings/contexts/<file>.md`). The `.agent-src.uncompressed/` prefix is rejected by the regex; the rewriter (`scripts/compress.py::_rewrite_paths`) resolves logical names at compress time. Rewritten relative forms (`../contexts/...`, `../../contexts/...`) are accepted so the linter passes on the compressed mirror in CI."
       },
       "description": "Lazy on-demand context references. Use logical names rooted at the source (e.g. `contexts/execution/foo.md`); the `.agent-src.uncompressed/` prefix is forbidden by the regex (road-to-path-fixes.md P5.3). Path rules and budget caps enforced by scripts/lint_load_context.py. Contract: docs/contracts/load-context-schema.md."
     },
@@ -38,7 +38,7 @@
       "type": "array",
       "items": {
         "type": "string",
-        "pattern": "^((\\.\\./)*contexts/|agents/contexts/|\\.agent-src/contexts/)[^\\s]+\\.md$",
+        "pattern": "^((\\.\\./)*contexts/|agents/settings/contexts/|\\.agent-src/contexts/)[^\\s]+\\.md$",
         "description": "Same logical-name rule as `load_context`."
       },
       "description": "Eager auto-loaded context references. Same logical-name rule as `load_context`. Counts against the per-rule char budget; enforced by scripts/lint_load_context.py."
@@ -98,6 +98,46 @@
         }
       },
       "description": "Per-rule allowlist consumed by the post-compression validator (scripts/check_compressed_paths.py). Rules that document forbidden path substrings as their subject matter (e.g. augment-portability, no-roadmap-references) declare the literal strings here so the gate does not flag itself. road-to-path-fixes.md P5.1."
+    },
+    "workspaces": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "pattern": "^[a-z][a-z0-9-]*$"},
+      "description": "ADR-013 workspace ids (additive discovery axis). Closed vocabulary enforced by scripts/lint_artefact_frontmatter.py against config/discovery/workspaces.yml."
+    },
+    "packs": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "pattern": "^[a-z][a-z0-9-]*$"},
+      "description": "ADR-013 pack ids (additive discovery axis). Closed vocabulary enforced by scripts/lint_artefact_frontmatter.py against config/discovery/packs.yml."
+    },
+    "lifecycle": {
+      "type": "string",
+      "enum": ["active", "deprecated", "experimental", "archived"],
+      "description": "ADR-013 lifecycle state."
+    },
+    "trust": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["level", "confidence", "human_review_required"],
+      "properties": {
+        "level": {"type": "string", "enum": ["core", "professional", "experimental", "advisory", "restricted"]},
+        "confidence": {"type": "string", "enum": ["high", "medium", "low"]},
+        "human_review_required": {"type": "boolean"}
+      },
+      "description": "ADR-013 trust block. Vocabulary mirrored by scripts/lint_artefact_frontmatter.py."
+    },
+    "install": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["default", "removable"],
+      "properties": {
+        "default": {"type": "boolean"},
+        "removable": {"type": "boolean"}
+      },
+      "description": "ADR-013 install hints (default-on at consumer install; user-removable)."
     }
   }
 }

package/scripts/schemas/skill.schema.json

@@ -2,7 +2,7 @@
   "$schema": "http://json-schema.org/draft-07/schema#",
   "$id": "https://github.com/event4u-app/agent-config/scripts/schemas/skill.schema.json",
   "title": "Skill frontmatter",
-  "$comment": "Source: agents/docs/frontmatter-contract.md#skills. Keep in sync with inventory_frontmatter.py and the linter's lint_skill().",
+  "$comment": "Source: agents/reference/docs/frontmatter-contract.md#skills. Keep in sync with inventory_frontmatter.py and the linter's lint_skill().",
   "type": "object",
   "additionalProperties": false,
   "required": ["name", "description", "source", "domain"],
@@ -35,6 +35,11 @@
       "type": "string",
       "pattern": "^[a-z][a-z0-9-]*$"
     },
+    "framework": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9-]*$",
+      "description": "Framework carve-out marker. Set on skills that are 100% coupled to one stack (e.g. `framework: laravel` on laravel-* skills). Generic, multi-stack skills MUST omit this key. Routed by .agent-src.uncompressed/rules/framework-neutrality-in-generic-skills.md (Tier 2)."
+    },
     "personas": {
       "type": "array",
       "items": {
@@ -117,6 +122,46 @@
           "items": { "type": "string" }
         }
       }
+    },
+    "workspaces": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "pattern": "^[a-z][a-z0-9-]*$"},
+      "description": "ADR-013 workspace ids (additive discovery axis). Closed vocabulary enforced by scripts/lint_artefact_frontmatter.py against config/discovery/workspaces.yml."
+    },
+    "packs": {
+      "type": "array",
+      "minItems": 1,
+      "uniqueItems": true,
+      "items": {"type": "string", "pattern": "^[a-z][a-z0-9-]*$"},
+      "description": "ADR-013 pack ids (additive discovery axis). Closed vocabulary enforced by scripts/lint_artefact_frontmatter.py against config/discovery/packs.yml."
+    },
+    "lifecycle": {
+      "type": "string",
+      "enum": ["active", "deprecated", "experimental", "archived"],
+      "description": "ADR-013 lifecycle state."
+    },
+    "trust": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["level", "confidence", "human_review_required"],
+      "properties": {
+        "level": {"type": "string", "enum": ["core", "professional", "experimental", "advisory", "restricted"]},
+        "confidence": {"type": "string", "enum": ["high", "medium", "low"]},
+        "human_review_required": {"type": "boolean"}
+      },
+      "description": "ADR-013 trust block. Vocabulary mirrored by scripts/lint_artefact_frontmatter.py."
+    },
+    "install": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["default", "removable"],
+      "properties": {
+        "default": {"type": "boolean"},
+        "removable": {"type": "boolean"}
+      },
+      "description": "ADR-013 install hints (default-on at consumer install; user-removable)."
     }
   }
 }

package/scripts/score_skill_selection.py

@@ -18,7 +18,7 @@ Predictions source:
   pure description-matching would do and provides a numeric floor.
 
 Output: human-readable summary on stdout + machine JSON to
-`agents/reports/skill-selection-accuracy.json` (or `--out`).
+`agents/runtime/reports/skill-selection-accuracy.json` (or `--out`).
 """
 
 from __future__ import annotations

package/scripts/skill_collision_clusters.py

@@ -7,7 +7,7 @@ high-overlap skill pairs into clusters. The output drives the
 selection-accuracy fixture set defined by council file 05 (Round-3
 protocol — ≥ 3 shared significant terms → collision cluster).
 
-Output: `agents/reports/skill-collision-clusters.json`
+Output: `agents/runtime/reports/skill-collision-clusters.json`
 
 Schema:
     {

package/scripts/skill_linter.py

@@ -403,7 +403,7 @@ def _fenced_content_ratio(text: str) -> float:
 # --- Structural-density model (docs/contracts/linter-structural-model.md) ---
 # Replaces the raw line/word/fenced-ratio gates with four primitives that
 # distinguish complexity from bloat. Calibrated 2026-05-08 against the full
-# 310-artefact corpus (agents/.density-snapshot.jsonl).
+# 310-artefact corpus (agents/runtime/density/snapshot.jsonl).
 
 PROCEDURE_HEADING_PATTERN = re.compile(
     r"^##\s+Procedure(\s*[:\u2014\-].*)?\s*$", re.MULTILINE
@@ -472,6 +472,21 @@ def _command_delegation_signal(text: str, frontmatter: Optional[str]) -> bool:
     return False
 
 
+def _strip_markdown_for_check(text: str) -> str:
+    """Strip fenced code, inline code spans, and markdown links so heuristic
+    regex matches operate on prose only.
+
+    Used by rule-body heuristics whose targets (e.g. ``procedural_rule``)
+    must not flip on legitimate skill pointers like ``[git-workflow](…)``
+    or ``` `skill:symfony-workflow` ```. Frontmatter is handled by the
+    caller via ``text.split("---", 2)[-1]``.
+    """
+    text = re.sub(r"```[^\n]*\n.*?```", "", text, flags=re.DOTALL)
+    text = re.sub(r"`[^`\n]+`", "", text)
+    text = re.sub(r"\[[^\]]*\]\([^)]*\)", "", text)
+    return text
+
+
 def _iron_law_blocks(text: str) -> int:
     """Count fenced blocks that look like verbatim Iron-Law imperatives.
 
@@ -575,10 +590,36 @@ def has_validation_step(procedure_block: str) -> bool:
     return any(signal in lowered for signal in good_signals)
 
 
+_INSPECT_VERB_PATTERN = re.compile(
+    r"\b(?:"
+    # Direct inspection
+    r"inspect|examine|audit|survey"
+    # Read / look
+    r"|read|look\s+at"
+    # Check (word-boundary — matches "check that", "check current", "check what")
+    r"|check"
+    # Review (broad — matches "review existing", "review the failures")
+    r"|review"
+    # Comprehension / orientation
+    r"|understand|identify|analyze|analyse"
+    # Discovery
+    r"|detect|gather|discover"
+    r")\b",
+    re.IGNORECASE,
+)
+
+
 def has_inspect_step(procedure_block: str) -> bool:
-    lowered = procedure_block.lower()
-    inspect_signals = ["inspect", "check current", "review existing", "identify", "analyze"]
-    return any(signal in lowered for signal in inspect_signals)
+    """Return True if the procedure block opens with an inspect / read step.
+
+    Corpus-driven verb list (see docs/contracts/linter-structural-model.md):
+    the first ordered step in a skill procedure should orient the agent in
+    the live system — read existing code, examine current state, detect
+    stack — before mutating anything. Regex uses word boundaries to avoid
+    substring matches inside unrelated words (e.g. ``read`` inside
+    ``already``).
+    """
+    return bool(_INSPECT_VERB_PATTERN.search(procedure_block))
 
 
 def find_vague_validation(text: str) -> list[str]:
@@ -678,8 +719,9 @@ def lint_skill(path: Path, text: str) -> LintResult:
     if skill_name and "-" not in skill_name and len(skill_name) >= 3:
         # Single word without qualifier — likely too generic
         ALLOWED_BARE_NOUNS = {"database", "devcontainer", "docker", "eloquent", "flux", "forecasting",
-                              "grafana", "laravel", "livewire", "mcp", "openapi", "performance",
-                              "security", "terraform", "terragrunt", "traefik", "websocket"}
+                              "grafana", "laravel", "livewire", "markitdown", "mcp", "openapi",
+                              "performance", "security", "terraform", "terragrunt", "traefik",
+                              "websocket"}
         if skill_name.lower() not in ALLOWED_BARE_NOUNS:
             issues.append(Issue("warning", "bare_noun_name",
                                 f"Bare-noun skill name `{skill_name}` — consider adding a qualifier (e.g., `{skill_name}-management`)"))
@@ -1504,9 +1546,20 @@ def lint_rule(path: Path, text: str) -> LintResult:
         if bad_sign in text:
             issues.append(Issue("error", "rule_looks_like_skill", f"Rule contains skill-like section: {bad_sign}"))
 
-    # Exclude frontmatter from procedural check (frontmatter may contain "type")
+    # Procedural-rule heuristic: a rule "looks procedural" only when its own
+    # prose AND its own structure both signal a procedure. We:
+    #   1. Exclude frontmatter (may contain "type", path strings, etc.).
+    #   2. Strip code spans, fenced blocks, and markdown links — so legitimate
+    #      pointers to procedural skills (e.g. `skill:git-workflow`,
+    #      [symfony-workflow](…)) do not flip the keyword count.
+    #   3. Require ≥ 2 keyword occurrences in stripped prose AND ≥ 3 ordered
+    #      steps AND no Iron-Law block — that combination distinguishes a
+    #      mis-classified procedure from a rule that merely references one.
     body = text.split("---", 2)[-1] if frontmatter else text
-    if re.search(r"\b(procedure|workflow)\b", body, re.IGNORECASE):
+    stripped_body = _strip_markdown_for_check(body)
+    kw_count = len(re.findall(r"\b(procedure|workflow)\b", stripped_body, re.IGNORECASE))
+    ordered_steps = len(re.findall(r"^\s*\d+\.\s+", body, re.MULTILINE))
+    if kw_count >= 2 and ordered_steps >= 3 and _iron_law_blocks(text) == 0:
         issues.append(Issue("warning", "procedural_rule", "Rule looks procedural; consider a skill instead"))
 
     return LintResult(
@@ -2919,7 +2972,7 @@ def lint_frontmatter_schema(path: Path, text: str, artifact_type: str) -> List[I
     """Validate the frontmatter of an artefact against its JSON-Schema.
 
     Schemas live in ``scripts/schemas/``. One schema per artefact type;
-    see ``agents/docs/frontmatter-contract.md`` for the human-readable
+    see ``agents/reference/docs/frontmatter-contract.md`` for the human-readable
     contract the schemas encode. Guidelines have no frontmatter and are
     skipped.
     """

package/scripts/skill_overlap.py

@@ -4,7 +4,7 @@
 Implements step-2-skill-inventory-rationalization.md Phase 2 Step 2.
 Mirrors `scripts/audit_overlap.py` (the rule-side analog) but reads
 `.agent-src.uncompressed/skills/<slug>/SKILL.md` frontmatter directly
-and emits `agents/metrics/skill-overlap.md` listing pairs scoring
+and emits `agents/runtime/metrics/skill-overlap.md` listing pairs scoring
 ≥ 0.6 on either:
 
   - description-trigger Jaccard (tokenized union of `description` +

package/scripts/skill_tools/run_block_d_eval.py

@@ -3,7 +3,7 @@
 
 Runs D2 (`score_skill_relevance`), D3 (`audit_persona_coverage`), and
 D4 (`suggest_skill_for_task`) against the corpora in
-`agents/eval-corpora/block-d/` and emits a pass/fail summary per the
+`agents/evidence/eval-corpora/block-d/` and emits a pass/fail summary per the
 council verdict targets:
 
   - **D2**: ≥ 85 % of corpus tasks have an `expected_top3` skill in

package/scripts/skill_usage_collect.py

@@ -14,12 +14,12 @@ parses each turn for two signals:
     `.agent-src/skills/<slug>/`.
 
 Emits one JSONL record per (session, turn, slug, kind) to
-`agents/metrics/skill-usage.jsonl` (append-only, deduped on the
+`agents/runtime/metrics/skill-usage.jsonl` (append-only, deduped on the
 (session_id, turn_idx, slug, kind) tuple).
 
 Privacy: `prompt_excerpt_hash` = SHA-256 of the first 200 chars of the
 user prompt that opened the turn. No raw user or assistant bodies are
-persisted. See `agents/audit-2026-05-14-north-star/skill-usage-sources.md`.
+persisted. See `agents/evidence/audits/2026-05-14-north-star/skill-usage-sources.md`.
 """
 from __future__ import annotations
 
@@ -153,7 +153,7 @@ def dedup_key(rec: dict) -> tuple:
 def main() -> int:
     ap = argparse.ArgumentParser(description=__doc__.splitlines()[0])
     ap.add_argument("--project-slug", help="Override the ~/.claude/projects slug (defaults to current repo)")
-    ap.add_argument("--out", type=Path, default=OUT, help="Output jsonl (default: agents/metrics/skill-usage.jsonl)")
+    ap.add_argument("--out", type=Path, default=OUT, help="Output jsonl (default: agents/runtime/metrics/skill-usage.jsonl)")
     ap.add_argument("--quiet", action="store_true", help="Suppress non-error output")
     args = ap.parse_args()
 

package/scripts/skill_usage_report.py

@@ -1,8 +1,8 @@
 #!/usr/bin/env python3
-"""Aggregate `agents/metrics/skill-usage.jsonl` into a per-skill report.
+"""Aggregate `agents/runtime/metrics/skill-usage.jsonl` into a per-skill report.
 
 Implements step-2-skill-inventory-rationalization.md Phase 1 Step 3.
-Groups records by slug; emits `agents/metrics/skill-usage-report.md`
+Groups records by slug; emits `agents/runtime/metrics/skill-usage-report.md`
 with columns:
 
   slug · exposures_total · mentions_total · exposures_30d · mentions_30d
@@ -112,7 +112,7 @@ def render(per: dict[str, dict], known: set[str]) -> str:
         "# Skill Usage Report (baseline)",
         "",
         "> Generated by `scripts/skill_usage_report.py`. Source:",
-        "> `agents/metrics/skill-usage.jsonl` (collector emits per-turn",
+        "> `agents/runtime/metrics/skill-usage.jsonl` (collector emits per-turn",
         "> exposure/mention records). See",
         "> [`step-2-skill-inventory-rationalization.md`](../roadmaps/step-2-skill-inventory-rationalization.md)",
         "> Phase 1.",

package/scripts/smoke/kernel.sh

@@ -2,9 +2,9 @@
 # scripts/smoke/kernel.sh — kernel-tier smoke (step-11 Phase 3 Step 2).
 #
 # Asserts:
-#   1. router.json lists exactly 9 kernel rules.
+#   1. router.json lists exactly 10 kernel rules.
 #   2. Every kernel rule file exists at .agent-src/rules/<id>.md.
-#   3. 8 of 9 carry at least one Iron-Law fenced block.
+#   3. 9 of 10 carry at least one Iron-Law fenced block.
 #      agent-authority is the dispatch index, exempt from the fence
 #      requirement (docs/contracts/smoke-contracts.md § 3.1).
 #   4. Kernel-bucket char budget breaches ≤ EXPECTED_BREACHES.
@@ -16,8 +16,8 @@
 
 set -euo pipefail
 
-EXPECTED_KERNEL_COUNT=9
-EXPECTED_FENCE_CARRIERS=8
+EXPECTED_KERNEL_COUNT=10
+EXPECTED_FENCE_CARRIERS=9
 EXPECTED_BREACHES=2
 EXEMPT_FROM_FENCE="agent-authority"
 

package/scripts/smoke/router.sh

@@ -2,7 +2,7 @@
 # scripts/smoke/router.sh — router-tier smoke (step-11 Phase 3 Step 3).
 #
 # Asserts router.json structural integrity:
-#   1. 75 ids = 9 kernel + 24 tier_1 + 42 tier_2 (locked count).
+#   1. 68 ids = 10 kernel + 23 tier_1 + 35 tier_2 (locked count).
 #   2. Every id resolves to .agent-src/rules/<id>.md (0 broken).
 #   3. Every routes_to ref resolves through its prefix
 #      (skill:, command:, guideline:, contract:); missing-contract
@@ -15,7 +15,7 @@
 
 set -euo pipefail
 
-EXPECTED_TOTAL_IDS=75
+EXPECTED_TOTAL_IDS=68
 EXPECTED_MISSING_CONTRACTS=2
 
 quiet="${SMOKE_QUIET:-0}"

package/scripts/smoke_quickstart.py

@@ -7,7 +7,7 @@ Verifies the 3-step Quickstart from a fresh-project perspective:
      `.agent-settings.yml` with the documented default `cost_profile`.
   2. The decision_engine block (P2.x of road-to-productization) parses
      cleanly through the same engine parser the runtime uses.
-  3. The work-engine state-file format (`agents/state/<id>.json`) is
+  3. The work-engine state-file format (`agents/runtime/state/<id>.json`) is
      emit-ready — schema for `decision_result` matches the contract.
 
 What it does NOT do:

package/scripts/spotcheck_thin_root.py

@@ -4,7 +4,7 @@
 Sends the refactored package-root AGENTS.md and the consumer template
 to Sonnet 4.5 + gpt-4o, asks each member to answer five questions
 that simulate a fresh agent landing on the file. Records qualitative
-verdicts in agents/reports/thin-root-platform-spotcheck.md.
+verdicts in agents/runtime/reports/thin-root-platform-spotcheck.md.
 """
 from __future__ import annotations