Builds the false-positive/true-positive proof the plan requires before the guard can be
promoted to blocking, and fixes the one false-positive it surfaced.
- test-harness-guard.sh: 12-case matrix in a throwaway repo, runs the REAL guard, asserts
WARN/clean for real conflicts/secrets/keys vs legit content (setext underlines, dividers,
docs that mention a marker, encrypted sops, public keys, .example templates).
- harness-guard.sh: conflict rule now requires a real hunk (BOTH ^<<<<<<< AND ^>>>>>>>),
dropping the lone =======$ trigger that false-positived on a 7-char setext underline /
divider. Identical true-positive power (git writes all three markers); FP surface -> 0.
- /self-check: new harness.guard_selftest runs the matrix in an isolated temp repo (read-only
vs the real tree) so guard correctness is continuously proven.
Verified 12/12 pass, true positives intact, real-tree FP surface = 0. FATAL flip (todo
f1c11d0d, on/after 2026-06-22) is now evidence-backed + one-step.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Task 3 leftover. Adds a 'consistency' category to /self-check that catches a standard
drifting back into restating/contradicting the command that owns the rule -- the Syncro
timers failure mode (standard said 'always timer' while /syncro said 'outlier only').
Deterministic half: each manifest.command_standard_links pair's standard must still carry
its defer-to-SSOT pointer (must_reference regex). Lost pointer = WARN. Seeded with
syncro-billing (time-entry-protocol.md -> /syncro). Semantic contradiction pass delegated
to the model in SKILL.md, mirroring check_memory. Verified PASS; negative-tested.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds a 'harness' category to /self-check (Task 12, self-check half) so the harness-
optimization gains can't silently regress. All read-only / non-invasive:
- VERSION marker present + not older than manifest.harness.min_version
- skill-registry description budget (sum of all SKILL.md description: fields under
registry_desc_budget_chars) -- the metric that catches Task 5 bloating back
- global deploy targets ~/.claude/skills + ~/.claude/commands populated (Mac-wipe failure)
- harness-guard.sh present + wired into sync.sh
- core scripts parse (bash -n on sync/guard/now-phoenix); now-phoenix.sh emits a valid date
Tunables in baseline/manifest.json 'harness' block. Verified 9/9 PASS; budget WARN
negative-tested at a synthetic over-budget value.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New /self-check skill: each machine probes its own ClaudeTools harness wiring
(identity.json paths, required tooling, settings.json hooks, skill/command/script
set, vault decrypt, coord/Gitea connectivity, Ollama capability tier) and grades
RED/AMBER/GREEN against a checked-in provisional baseline manifest.
- Capability-tier model: architectural/OS/hardware differences (e.g. no local
Ollama) select a fallback ruleset instead of failing.
- Duplicate detection: flags command/skill names that diverge between the repo
and ~/.claude (the "same /cmd, different behaviour" cross-machine bug);
CRLF-only diffs ignored.
- Memory check: index + orphan detection, plus a model-driven semantic pass for
memories that contradict identity/settings.
- V1 is a census tool: --publish writes a per-machine census to coord
(component selfcheck_<host>); fanout requests the fleet to self-check +
self-remediate + re-publish; aggregate derives the proposed baseline. No
machine ever fixes another.
Reviewed twice by the Code Review Agent; three CRITICAL coord-API bugs and the
CRLF false-WARN found and fixed, verified live against the coord API.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
