TalonT-Org · Trecek · Jul 25, 2026 · Jul 23, 2026 · Jul 23, 2026 · Jul 23, 2026
diff --git a/.autoskillit/skills/audit-feature-gates/SKILL.md b/.autoskillit/skills/audit-feature-gates/SKILL.md
@@ -1,6 +1,7 @@
 ---
 name: audit-feature-gates
 categories: [audit]
+uses_capabilities: [agent_model]
 description: >
   Audit feature flag isolation — traces import chains, runtime gates, tool/skill
   tag coverage, UI surfaces, and test markers to detect leakage and miswiring.

diff --git a/.autoskillit/skills/eval-agent/SKILL.md b/.autoskillit/skills/eval-agent/SKILL.md
@@ -1,6 +1,7 @@
 ---
 name: eval-agent
 categories: [eval]
+uses_capabilities: [agent_subagent]
 description: >
   Invoke a named agent definition against a provided prompt and capture its output.
   Execution primitive for the agent-eval recipe — isolates a single agent invocation.

diff --git a/.claude/skills/audit-bugs/SKILL.md b/.claude/skills/audit-bugs/SKILL.md
@@ -1,5 +1,6 @@
 ---
 name: audit-bugs
+uses_capabilities: [claude_dir]
 description: Analyze historical bug patterns by mining Claude Code project logs for /investigate skill invocations since a specified date. Identifies recurring root causes, architectural gaps, and proactive detection strategies. Use when user says "audit bugs", "bug patterns", "analyze investigations", or "bug audit".
 hooks:
   PreToolUse:

diff --git a/.claude/skills/chart-course/SKILL.md b/.claude/skills/chart-course/SKILL.md
@@ -1,5 +1,6 @@
 ---
 name: chart-course
+uses_capabilities: [agent_model, cross_skill_ref]
 description: Interactive strategic compass builder. Guides the user through mapping all possible project directions with progressive codebase analysis, web research, and architectural diagrams at every step. Produces a machine-readable compass document for downstream alignment tracking.
 hooks:
   PreToolUse:

diff --git a/.claude/skills/check-bearing/SKILL.md b/.claude/skills/check-bearing/SKILL.md
@@ -1,5 +1,6 @@
 ---
 name: check-bearing
+uses_capabilities: [agent_model]
 description: Assess a branch or PR's alignment with the strategic compass. Evaluates whether changes advance, drift from, or close off strategic directions. Produces an alignment dashboard with per-direction impact analysis and a verdict.
 hooks:
   PreToolUse:

diff --git a/.claude/skills/promote-to-main/SKILL.md b/.claude/skills/promote-to-main/SKILL.md
@@ -1,6 +1,7 @@
 ---
 name: promote-to-main
 categories: [github]
+uses_capabilities: [agent_model, cross_skill_ref, github_api_write]
 description: >
   Promote integration to main with comprehensive changelog and PR creation. Use when
   user says "promote to main", "open promotion PR", "integration to main", or
@@ -96,10 +97,7 @@ working directory.
 ```bash
 mkdir -p .autoskillit/temp/promote-to-main
 python3 - <<'EOF' > .autoskillit/temp/promote-to-main/token_summary.md 2>/dev/null || true
-import json, pathlib, sys
-from autoskillit.pipeline.tokens import DefaultTokenLog
-from autoskillit.pipeline.telemetry_fmt import TelemetryFormatter
-from autoskillit.execution.session_log import resolve_log_dir
+import json, os, pathlib, sys
 
 cfg_path = pathlib.Path(".autoskillit") / "temp" / ".hook_config.json"
 kitchen_id = ""
@@ -108,14 +106,41 @@ if cfg_path.exists():
     if isinstance(_cfg, dict):
         kitchen_id = _cfg.get("kitchen_id") or _cfg.get("pipeline_id", "")
 
-log_root = resolve_log_dir("")
-tl = DefaultTokenLog()
-n = tl.load_from_log_dir(log_root, kitchen_id_filter=kitchen_id)
-if n == 0:
+log_root = os.environ.get("AUTOSKILLIT_LOG_DIR", "")
+if not log_root:
     sys.exit(0)
-steps = tl.get_report()
-total = tl.compute_total()
-print(TelemetryFormatter.format_token_table(steps, total))
+tl_path = pathlib.Path(log_root)
+if not tl_path.exists():
+    sys.exit(0)
+
+total_input = total_output = 0
+session_count = 0
+for session_dir in sorted(tl_path.iterdir()):
+    if not session_dir.is_dir():
+        continue
+    token_file = session_dir / "tokens.json"
+    if not token_file.exists():
+        continue
+    try:
+        data = json.loads(token_file.read_text())
+        if kitchen_id and data.get("kitchen_id") != kitchen_id:
+            continue
+        session_count += 1
+        total_input += data.get("input_tokens", 0)
+        total_output += data.get("output_tokens", 0)
+    except (json.JSONDecodeError, OSError, TypeError):
+        continue
+
+if session_count == 0:
+    sys.exit(0)
+
+header = "| Metric | Value |\n|---|---|\n"
+rows = (
+    f"| Sessions | {session_count} |\n"
+    f"| Input Tokens | {total_input:,} |\n"
+    f"| Output Tokens | {total_output:,} |\n"
+)
+print(header + rows)
 EOF
 ```
 

diff --git a/.claude/skills/review-promotion/SKILL.md b/.claude/skills/review-promotion/SKILL.md
@@ -1,6 +1,7 @@
 ---
 name: review-promotion
 categories: [github]
+uses_capabilities: [agent_model]
 description: >
   Reviewer-facing deep analysis of an integration-to-main promotion. Performs domain
   risk scoring, breaking change audit, regression risk assessment, test coverage delta,

diff --git a/.claude/skills/validate-audit/SKILL.md b/.claude/skills/validate-audit/SKILL.md
@@ -1,6 +1,7 @@
 ---
 name: validate-audit
 categories: [audit]
+uses_capabilities: [agent_model, github_api_write]
 description: Validate audit findings from audit-arch, audit-tests, or audit-cohesion against actual code, git history, and design intent using 9–10 parallel subagents. Removes contested findings, documents exceptions, adjusts severities. Use when user says "validate audit", "validate findings", "validate report", or "check audit results".
 hooks:
   PreToolUse:

diff --git a/docs/README.md b/docs/README.md
@@ -35,4 +35,5 @@ multi-level orchestrator. The bundled recipes implement issue → plan → workt
 - [research/audit-trail-format.md](research/audit-trail-format.md) — audit/ artifact structure and lifecycle
 - [research/codex-delivery-conformance.md](research/codex-delivery-conformance.md) — Codex recipe envelope/pull conformance and protected-host blocker
 - [audit/surface-freeze-checklist.md](audit/surface-freeze-checklist.md) — commands.py public import surface freeze checklist
+- [verification/review-pr-immunity.md](verification/review-pr-immunity.md) — deterministic review-pr projection matrix and live provider-attempt record
 - [phoropter/](phoropter/README.md) — phoropter lens framework: execution contracts, recipe blocks, synthesis strategies, and authoring guide
diff --git a/docs/configuration.md b/docs/configuration.md
@@ -334,9 +334,12 @@ skills:
     # ...
 ```
 
-Any bundled skill can be promoted or demoted by adding it to the desired tier list. A skill
-in multiple tiers simultaneously is a validation error. See **[Skill Visibility](skills/visibility.md)**
-for the full tier breakdown, session mode table, and override rules.
+Session-role skills can be promoted or demoted by adding them to the desired tier list. A
+skill in multiple tiers simultaneously is a validation error. Exact-role orchestration
+skills are not user-tiered: `process-issues`, for example, is exposed only in L2
+orchestrator catalogs and a configuration that adds it to a session tier is rejected.
+See **[Skill Visibility](skills/visibility.md)** for the full tier breakdown, session mode
+table, role-derived catalogs, and override rules.
 
 ## Subset Categories
 

diff --git a/docs/design/acp-session-contract.md b/docs/design/acp-session-contract.md
@@ -175,7 +175,7 @@ The contract nudge exclusively targets `session/resume`; it never invokes
 ## Section 3: Capabilities Translation
 
 `BackendCapabilities` (`src/autoskillit/core/types/_type_backend.py`,
-frozen dataclass, 41 fields total) declares feature flags the orchestrator
+frozen dataclass, 40 fields total) declares feature flags the orchestrator
 consumes when selecting an ACP rung or backend-specific code path. Each field
 falls into one of three categories:
 
@@ -186,7 +186,7 @@ falls into one of three categories:
   outside the exemption set. Membership is validated against `_FORWARD_DECLARED`
   in `tests/arch/test_capability_consumption.py`.
 
-The counts below are **17 ACP-Mappable + 6 Forward-Declared + 18 autoskillit-Local = 41 total**.
+The counts below are **17 ACP-Mappable + 6 Forward-Declared + 17 autoskillit-Local = 40 total**.
 
 ### 3.1 Category 1: ACP-Mappable (17 fields)
 
@@ -210,11 +210,10 @@ The counts below are **17 ACP-Mappable + 6 Forward-Declared + 18 autoskillit-Loc
 | `record_capable` | ACP scenario recording |
 | `inspector_capable` | ACP health monitoring callback (Health Inspector per issue #3533) |
 
-### 3.2 Category 2: autoskillit-Local Extension (18 fields)
+### 3.2 Category 2: autoskillit-Local Extension (17 fields)
 
 | Field | autoskillit-specific contract |
 |---|---|
-| `project_local_skills_capable` | autoskillit `.claude/skills/` discovery (Claude: `True`; Codex: `False`) |
 | `supports_tool_list_changed` | autoskillit kitchen reveal timing (tool list notification) |
 | `required_skill_fields` | autoskillit `SKILL.md` validation (`frozenset({"name", "description"})`) |
 | `applicable_guards` | autoskillit guard script enforcement (Claude: `{"skill_load_guard"}`; Codex: `{"write_guard"}`) |
@@ -362,7 +361,7 @@ warning (the others are static no-ops with no observable behavior).
 | §2 RetryReason enum | `RetryReason` | `src/autoskillit/core/types/_type_enums.py` lines 44–64 |
 | §2 Retry routing | `_compute_retry`, `_build_skill_result` overrides | `src/autoskillit/execution/session/_retry_fsm.py`, `src/autoskillit/execution/headless/_headless_result.py` |
 | §2 Contract nudge | `_attempt_contract_nudge`, `_merge_token_usage` | `src/autoskillit/execution/headless/_headless_recovery.py` |
-| §3 Capabilities | `BackendCapabilities` (41 fields) | `src/autoskillit/core/types/_type_backend.py` |
+| §3 Capabilities | `BackendCapabilities` (40 fields) | `src/autoskillit/core/types/_type_backend.py` |
 | §3 Forward-declared | `_FORWARD_DECLARED` | `tests/arch/test_capability_consumption.py` |
 | §4 Codex flags | `CodexFlags` | `src/autoskillit/execution/backends/codex.py` lines 98–107 |
-| §4 Codex discard sites | `codex.py` F841 / warning sites | `src/autoskillit/execution/backends/codex.py` |
+| §4 Codex discard sites | `codex.py` F841 / warning sites | `src/autoskillit/execution/backends/codex.py` |