- Fall back to the first claimed repository when the preferred canonical root has no paths, so plans that only remove installs under ~/.agents still mark a canonical entry
Agent-Session: claude/b488c619-f5a4-48e1-8cd5-c640564fc26e
- Emit plan --json schema v2 with the head SHA and per-repo ai-coord claim descriptors so the publish runbook no longer derives claim paths by hand
- Replace the selected/candidateCount/driftCount JSON fields with selectedCount and filters
- Validate that SKILL.md frontmatter name matches the skill directory
- Add readme-skills-check script and just recipe verifying README skills tables against skills/
Agent-Session: claude/b488c619-f5a4-48e1-8cd5-c640564fc26e
Type helper-script inputs and domain models so the inherited Bun strict and unchecked-index settings pass without changing CLI behavior.
Agent-Session: codex/019fd280-c0b1-7f50-a0e6-0c7aea28e10c
- Route evidence and catalog discovery through the routemesh CLI
- Preserve checkpoint and coverage safeguards
Agent-Session: codex/019fcbc9-fc31-7772-9bc9-1c7af36775f5
- Reject the unsafe FTMScout route while preserving its Chainscout provenance.
- Route Opera account evidence to the official GraphQL endpoint with explicit partial-index and internal-history boundaries.
- Regenerate the canonical Fantom explorer URL from crypto-registry.
Group references/ by provenance and topic: generated outputs move to
references/generated/, workflow playbooks to references/workflows/,
explorer docs to references/explorers/, and bridge docs to
references/bridges/ (bridge- prefix dropped). Rename
optimism-pre-2021-11-11.md to optimism-pre-regenesis.md.
Standardize all intra-skill links on skill-root-relative paths without
a ./ prefix, route SKILL.md to the previously unreachable Etherscan and
Blockscout API docs, and replace hand-copied chain and plan tables with
pointers to their single owning files. Update generate-evm-atlas.mjs
output paths and embedded link strings to match.
- Pin address sweeps to a fixed finalized/verified checkpoint instead of `latest`, batching nonce/balance RPC calls and bounding indexed history queries to that cutoff
- Gate the zero-nonce/zero-balance shortcut on `accountActivityModel`, defaulting to deny for account-abstraction, cross-VM, or unknown chains
- Add `general` and `prb-finance bootstrap` coverage profiles with distinct required channels
- Add optional multi-provider quorum for positive/negative results
- Add `accountActivityModel` to every target-mainnets.json chain and validate it in generate-evm-atlas.mjs against crypto-registry schema v2
- Add fixtures and validation scripts for Blockscout address counters and Etherscan transfer-topic log queries
- Add a --discover-routemesh mode to scripts/generate-evm-atlas.mjs that checks atlas-overlays.json's routeMesh claims against RouteMesh's live chain list, instead of trusting hand-typed booleans that can go stale.
- Correct routeMesh to true for Robinhood Chain and Zora, both incorrectly marked false.
- Add the just evm-atlas-discover-routemesh recipe and document it in AGENTS.md.
Load registry data directly from data/chains.json instead of importing the built dist/index.js, removing the need for a fresh tsc-build and the source-vs-dist mtime freshness check.
- Replace mdformat with pinned Prettier checks, formatting, and hooks
- Format generated EVM Atlas references through the shared Prettier config
- Reformat the Markdown baseline
- Add scripts/generate-evm-atlas.mjs to derive evm-atlas references from @prb/crypto-registry plus references/atlas-overlays.json
- Add just evm-atlas-generate/evm-atlas-check recipes and wire the check into lint-staged
- Regenerate chain-aliases.json, target-mainnets.json, blockscout-chains.md, and etherscan-chains.md from the new generator
- Update evm-atlas/scripts/resolve-chain.sh for the regenerated alias data
- restore SKILL.md invocation fields as the source of truth
- update Codex metadata to match disable-model-invocation
- relax the sync checker for Claude frontmatter defaults
Paul Berg