[sentinelhood]

Follow-up to #1244.

Adds replayable evidence receipts and reorg-aware monitoring:

• stores the last observed block hash;
• detects canonical-history changes;
• rewinds a safety window;
• replays observations before emitting an alert.

Validation:

• skill-pack validator: passed
• README/catalog validator: passed
• skill security scanner: passed
• embedded JSON validation: passed
• git diff --check: passed

The repository-wide eyebrowlock baseline drift is not asserted by this PR.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[aaronjmars]

nice follow-up. the #1244 review fixes (8h clamp, erc-721 split, log-limit halving) are good.

2 things before merge:

1. reorg replay says to drop alerted_tx hashes first seen inside the replay range, but alerted_tx only stores hashes, no block. the agent can't do that. store {hash, block} or drop that line.
2. the evidence receipt has one chain / chain_id / block_hash, but one run covers every watch across chains. needs one entry per chain (or one receipt per chain).

nit: memory/evidence/ gets a new file every run, forever. add a prune (keep 30 days?).

[sentinelhood]

Addressed the review points in commit ded2a980:

• alerted_tx entries now carry {hash, block_number, first_seen} so replay can remove entries inside the reorg window.
• Evidence receipts use one chains[] entry per chain, with chain-local ranges, sources, counters, and transaction hashes.
• Added 30-day pruning for evidence receipts.
• Rebased the follow-up onto current main after docs(ecosystem): drop dead ai2human.io link #1254.

Validation: skill security scan, category check, evidence JSON example parse, and git diff --check pass. The repository pack validator is unavailable here because this checkout has no skills-pack.json; this PR documents a skill contract and does not claim end-to-end runtime execution.

[Svector-anu]

ran the full local checklist against this branch: check-skill-categories, check-capabilities-parity, validate-config.js, validate-readme-catalog.mjs, generate-skills-json (ci's normalized diff), generate-skill-icons --check, lint-shell.sh, biome ci ., dashboard typecheck. all pass. eyebrowlock.json diff is just this skill's own hash updating, as expected for a skill.md change.

read through the reorg logic, the evidence receipt schema, and the erc-20/erc-721 disambiguation by topic count. internally consistent, atomic state writes, evidence explicitly scoped as an audit trail not a safety verdict. solid follow-up to #1244.

approving.