pool cap sweep: post each uncapped pool once, then only count it

[rbcp18]

Why

The nightly pool cap sweep posts the same ~40 uncapped pools on the same six chains to Slack every run. Most of them belong to other owners, and the list never changes, so after the first night the post is mostly noise.

What changes

In the AUDIT_NETWORKS sweep in packages/contracts/scripts/deploy-chain.sh:

• Remembers what it posted. Posted pools are kept in $AUDIT_STATE_DIR/pool-cap-audit-seen.txt, one <network> <pool> line each. AUDIT_STATE_DIR should point at a Railway volume on the scheduled service.
• New pools are posted once. A pool is named in Slack the first run it shows up uncapped, whoever owns it: address, pair, amount available and owner. After that it only counts toward a one-line total, e.g. Total: 38 uncapped pool(s) across 6 chain(s).
• The seen file tracks the current state. It is rewritten every run with the pools that are uncapped now, so a pool that gets capped and later uncapped counts as new again. A chain the run could not read keeps its previous lines.
• Unreadable chains are always posted, as one line with the reason.
• Quiet nights post nothing. With no new pools and no unreadable chains, the sweep prints nothing new since the last post; Slack skipped (...).
• The first run with a state file only records a baseline and posts the totals, so the channel doesn't get the whole list again.
• Without AUDIT_STATE_DIR, every uncapped pool is treated as new each run, in the shorter format.
• Exit codes are unchanged.

Testing

I ran the sweep block locally with a stub yarn that returns canned audit output and a stub curl that captures the Slack payload:

Run	Expected	Result
First run with a state dir	baseline and totals only	✓
Same pools again	nothing posted	✓
Same pools, one chain unreadable	"no new uncapped pools" plus a could-not-audit line	✓
One new pool added	that pool posted, with the total	✓
Same again	nothing posted	✓
Chain unreadable	its earlier seen entries are kept	✓
No AUDIT_STATE_DIR	every pool posted as new	✓

bash -n passes.

Rollout

After merge, bump CONTRACTS_REF in worker-agents/chain-deployer/Dockerfile. The pool-cap-audit service already has a volume at /data and AUDIT_STATE_DIR=/data set.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TEE2vGbGJPRMNNsZdNTZps

---

Generated by Claude Code

[rbcp18]

Closing without merging. Deciding what gets posted to Slack is internal ops work, not protocol code, so it's moving to the scheduled job's own repo. The audit-pool-caps task stays here, unchanged, and the new job calls it with --json true.

---

Generated by Claude Code