@chrono-meta/fh-gate 1.4.23 → 1.4.24

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@chrono-meta/fh-gate",
-  "version": "1.4.23",
+  "version": "1.4.24",
   "description": "FH runtime adapters — run FH governance, skills, and agents via Claude or Codex with machine-parseable gates.",
   "license": "MIT",
   "keywords": [

package/plugins/fh-meta/skills/goal-quench/SKILL.md

@@ -298,3 +298,14 @@ Phase 1: token-budget-gate verdict output + mode resolved (core default, or pro/
 ```
 
 Verdict: PASS (CLEAN verification) | CONDITIONAL_PASS (PENDING verification) | FAIL (BLOCKED verification) | ESCALATE (user decision required on partial completion)
+
+**Completion-claim discipline (Ralph — no silent partial completion):** a PASS/CONDITIONAL_PASS
+verdict must **enumerate, not assert** — (a) **evidence artifact**: the pipeline-conductor verification
+verdict + the calibration record path; (b) **failed/skipped checks by name** (which pipeline axes /
+sidecars FAILED or were skipped). This list is **not self-asserted** — reconcile it against the resolved
+mode's in-scope axis/sidecar set: a list shorter than (in-scope − passed) is incomplete, not PASS (an
+empty list is valid only when every in-scope check passed). This is the mechanical anchor that keeps the
+clause off a judge-only path; (c) **explicit residual risk beyond what the mode-qualifier already states**
+(e.g. an axis that ran but on degraded input). A verdict that omits any of the three is **incomplete, not
+PASS**. (Convergent sister evidence: oh-my-claudecode Ralph "refuses silent partial completion"; Hermes
+Agent `/goal` per-turn completion judge.)

package/plugins/fh-meta/skills/memory-hygiene/SKILL.md

@@ -121,6 +121,16 @@ echo "---\ndate: $(date +%Y-%m-%d)\nentries_checked: N\nupdated: N\ndrifted: N\n
 ## Constraints
 
 - **No auto-deletion**: Archive candidates are proposed, not deleted. Human confirmation required.
+- **Snapshot before archive (Destructive-Op Gate for memory)**: before applying any confirmed
+  archive/removal in Step 4, snapshot **every memory root the entry could live in** first — tar **all**
+  roots Step 1 enumerated (`~/.claude/projects/*/memory/` AND hub-local `memory/`), not just one:
+  `tar czf tracks/_meta/memory_snapshot_$(date +%Y%m%d-%H%M%S).tgz <each Step-1 root that exists>` — so
+  a wrong archive in any scanned root is one-command recoverable (a single-root snapshot can pass while
+  the archived entry lived in the other root — bind the scope to the entry, not to "a tarball exists").
+  Archive moves entries to a `.archive/` sibling, never hard-deletes (mirrors the Curator's
+  tar-before-pass + rollback discipline; the Step 2 staleness thresholds are already the mechanical
+  no-LLM pre-pass that the judged Step 3 re-verify sits on top of — order preserved: mechanical roster
+  first, judged re-verify second).
 - **Feedback type**: Never auto-modify feedback rules — flag only.
 - **Simplification guard**: If 0 stale entries found, output one line "memory-hygiene: all fresh" and exit.
 - **FH online advantage**: Unlike offline harnesses, FH can re-verify external references live.
@@ -130,9 +140,10 @@ echo "---\ndate: $(date +%Y-%m-%d)\nentries_checked: N\nupdated: N\ndrifted: N\n
 
 ```
 Step 1~5 complete
-+ Staleness roster output
++ Staleness roster output (Step 2 = mechanical no-LLM pre-pass)
 + Re-verification run for all STALE entries
 + User gate presented and responded to (y/N per item)
++ If any archive confirmed: snapshot covering the archived entry's root written before the move (tracks/_meta/memory_snapshot_*.tgz, spanning all existing Step-1 roots) — check class: mandatory-pass (snapshot covers the entry's dir, not merely "a tarball exists")
 + Hygiene log written to tracks/_meta/memory_hygiene_{date}.md
 ```
 

package/plugins/fh-meta/skills/pipeline-conductor/SKILL.md

@@ -268,3 +268,14 @@ Step 0 scope confirmed and scope translation table applied
 Verdict: PASS (all conditions met, sweep complete) | CONDITIONAL_PASS (sweep complete, pending items captured) | FAIL (chain halted, blocking items remain) | ESCALATE (chain paused, human decision required)
 
 A sweep is not done until the Step 5 report is output. Emitting per-step verdicts without the aggregated report is incomplete.
+
+**Completion-claim discipline (Ralph — no silent partial completion):** the Step 5 report's overall
+verdict must **enumerate, not assert** — (a) **evidence artifact**: the per-step verdict source (report
+path + the verdict lines it aggregates); (b) **failed + skipped checks listed by name** (degraded /
+skipped / `--quick`-omitted steps included). This list is **not self-asserted** — reconcile it against
+the Step 0 in-scope step set: a list shorter than (in-scope − passed) is incomplete (an empty list is
+valid only when every in-scope step passed). This is the mechanical anchor that keeps the clause off a
+judge-only path; (c) **explicit residual risk beyond what the mode-qualifier states** (e.g. a single-skill
+scope is not a full-harness claim). A CLEAN verdict that omits any of the three is **incomplete**.
+(Convergent sister evidence: oh-my-claudecode Ralph "refuses silent partial completion"; Hermes Agent
+`/goal` per-turn completion judge.)