manual-rule-liveness-prober
Probes a manifest manual rule for liveness via with/without-rule fresh-subagent dispatch, capturing a RED→GREEN delta (baseline fails → rule-loaded complies); session-bound, reporting-only, never CI.
manual-rule-liveness-prober
Status: shipped-beta (framework-side authoring tool) · Ships to: clone — excluded from consumer delivery (setup.d/20-agents.sh:26 case-skip, #552) · Fires at: operator-run in a framework session, top-level only (claude --agent, not a dispatched subagent — agents/manual-rule-liveness-prober.md:130); never wired into CI (:131)
What it is
Session-bound RED→GREEN liveness probe for manifest manual rules (check.type==='manual'): dispatch a fresh subagent without the rule (expect the observable-failure), then with the rule's policy text loaded (expect the observable-compliance), and report the delta (agents/manual-rule-liveness-prober.md:11). It closes the gap principle 02 leaves open — the mechanical gate proves structural liveness only; "Structure is the floor; behaviour is the proof" (agents/manual-rule-liveness-prober.md:18). Reporting-only: no edits to manifest, rule text, or source; the only artefact is a probe report (agents/manual-rule-liveness-prober.md:20).
How it works
- Why not CI. A fresh subagent per pass is an LLM dispatch on the operator's own subscription; no paid LLM call may run in CI, so the probe is session-bound and operator-initiated (
agents/manual-rule-liveness-prober.md:26). - Input. A single manual rule id or the keyword
all; the manual rules at time of writing areR10/R13/R18(code-grep-shaped, demoable) andIR5/IR6(runtime-shaped, behavioural demo deferred) (agents/manual-rule-liveness-prober.md:32-43). - Two-source scenario lookup. Pressure-scenarios read first from
.ai-factory/generated-scenarios.json(validated generated scenarios), falling back topackages/core/manifest/rules-manifest.json(agents/manual-rule-liveness-prober.md:47-50). - Demoability split. code-grep-shaped rules run the full RED→GREEN dispatch; runtime-shaped rules report structural-validation-only — "Do NOT run a text RED→GREEN on the code-marker and present it as proof of runtime liveness" (
agents/manual-rule-liveness-prober.md:66-69). - Verdicts. Exactly one of LIVE (the only verdict that proves liveness), BASELINE-DIDN'T-FAIL (T-V3-B — strengthen the baseline and re-run, never conclude redundancy from one non-failing baseline), WITH-RULE-DIDN'T-COMPLY, RUNTIME-SHAPED (DEFERRED) (
agents/manual-rule-liveness-prober.md:90-95).
Satellites & companions
- USES: the pressure-scenario contract (rules-manifest.schema.json + principle 02) and the manifest's fixed pressure vocabulary
time/authority/sunk-cost/scope-creep(agents/manual-rule-liveness-prober.md:57). - ADAPTS: Superpowers
writing-skillspressure-scenario methodology — "same RED→GREEN mechanism, different artifact": a SKILL that imparts a capability vs. a RULE that constrains a behaviour (agents/manual-rule-liveness-prober.md:125). - ADDS: the behavioural channel of two-channel liveness — principle 02 is the structural gate; running this prober on
R10/R13/R18is the behavioural proof (agents/manual-rule-liveness-prober.md:140-141). - Census satellite: B3. The two later probers (shipped-agent C19, getff-cold-run C20) reuse its fresh-subagent-per-pass machinery as their ADAPT source.
Anchors
agents/manual-rule-liveness-prober.md:2—name: manual-rule-liveness-proberagents/manual-rule-liveness-prober.md:3—description: Probes a manifest manual rule for liveness via with/without-rule fresh-subagent dispatch, capturing a RED→GREEN delta …setup.d/20-agents.sh:26—manual-rule-liveness-prober.md) continue ;; # authoring-only tool (#552)agents/manual-rule-liveness-prober.md:11—> **Authoritative for:** the `manual-rule-liveness-prober` sub-agent prompt — the session-bound RED→GREEN liveness probe for manifest manual rules …agents/manual-rule-liveness-prober.md:18—The point of this role: principle 02's mechanical gate proves a manual rule _has_ a well-formed pressure-scenario (structural liveness …agents/manual-rule-liveness-prober.md:45—## Step 1 — Load the rule's pressure-scenario + policy textagents/manual-rule-liveness-prober.md:66—## Step 2 — Classify demoability (do this BEFORE dispatching)agents/manual-rule-liveness-prober.md:88—## Step 5 — Compute and report the RED→GREEN deltaagents/manual-rule-liveness-prober.md:130—- **Top-level only.** Must be invoked via top-level `claude --agent`, not as a dispatched subagent (a normal CC subagent cannot spawn subagents). …
All anchors at framework pin b069c59328aa8e08671d3a19fd25efa618dfc1c5 (fetched 2026-09-11 from artyhoo/getff staging; ancestry of census pin a1337cb301 verified).
dual-channel-drift-auditor
Cold pairwise audit of a declared @dual-pair anchor group — measures verbatim overlap and divergence, and judges each group INTENTIONAL-TWIN / SSOT-POINTER / COPY-RISK / DRIFT; reporting-only, never CI-invoked.
shipped-agent-liveness-prober
Probes each shipped sub-agent for behavioural liveness via with/without-tools fresh-subagent dispatch, capturing a RED→GREEN delta (tool-less fabricates → tool-using cites real evidence); status DORMANT, operator-initiated only.