powerloom-aeon-skill

Canonical skill package for Aeon. Published as powerloom/aeon-skills on GitHub.

Do not edit the fork copy (anomit-aeon/skills/powerloom-bds) as source of truth. Change this repo, then re-install into your Aeon fork.

What makes this good

Whale Radar on Aeon is not “an LLM that polls Uniswap.” It is:

1. Verified block data — trades and USD amounts from BDS epoch snapshots; alerts include on-chain verification (CID, epoch) when present.
2. Deterministic catch-up — fetch-bds-epochs.py walks GET /mpp/snapshot/allTrades/{block} from cursor+1 to chain tip (up to 10 snapshots/run). No tip-only sampling, no ignored from_epoch query hacks.
3. OpenClaw parity, headless runtime — same whale logic as powerloom-bds-univ3/scripts/whale-cron.mjs, but split for GitHub Actions: prefetch owns fetch + cursor; postprocess-bds.sh dispatches ./notify (Claude only logs).
4. Rich, Telegram-safe alerts — pool metadata (USDC/WETH), BUY/SELL, Etherscan links; project slug only in verification footer.
5. Cheap agent step — skill reads .bds-cache/alerts.json for counts/logging only; no sandbox network, no LLM rewriting powerloom-bds-state.json.

Deep dive: powerloom-bds/references/architecture.md

Repo layout

powerloom-aeon-skill/
├── powerloom-bds/              # Skill tree (SKILL.md + references)
├── scripts/
│   ├── prefetch-bds.sh         # Orchestrator (wipes .bds-cache/, runs fetch + process)
│   ├── fetch-bds-epochs.py     # Per-block snapshot loop (recommended BDS API)
│   ├── bds_normalize.py          # Snapshot shape helpers
│   ├── process-bds-skill.py      # Whale threshold, dedupe, alert formatting, cursor
│   └── postprocess-bds.sh        # ./notify dispatch + cursor backup
├── templates/powerloom-bds.yml.example
├── install-into-aeon.sh
└── README.md

What stays in the Aeon fork

Path	Owner
aeon.yml	Enable schedule
memory/powerloom-bds-state.json	Runtime cursor (committed by Actions)
memory/powerloom-bds.yml	Operator threshold / mode
.bds-cache/	Ephemeral — must be in .gitignore
GitHub secret BDS_API_KEY	Your fork

Install

Full guide: INSTALL.md (blog-ready, any Aeon fork)

git clone https://github.com/powerloom/aeon-skills.git
cd /path/to/your-aeon-fork
/path/to/aeon-skills/install-into-aeon.sh

With no argument the script uses $(pwd) — cd into your fork first, or pass the fork path explicitly. Both are valid.

Copies skill + five scripts; runs patch-aeon-fork.sh for .gitignore, workflow BDS_API_KEY, and cursor auto-commit.

How it runs

messages.yml (cron) → aeon.yml skill=powerloom-bds
  → prefetch-bds.sh
      → fetch-bds-epochs.py   # snapshot/{block} loop
      → process-bds-skill.py  # alerts.json + state update
  → Claude reads SKILL.md → log alert count (no ./notify)
  → postprocess-bds.sh      # ./notify per alert (needs TELEGRAM_* on post-process env)
  → commit memory/powerloom-bds-state.json

Quick start

1. Fork Aeon, run install-into-aeon.sh
2. Secret BDS_API_KEY from https://bds-metering.powerloom.io/metering
3. Optional: cp templates/powerloom-bds.yml.example memory/powerloom-bds.yml
4. Enable in aeon.yml (e.g. */5 * * * * — effective dispatch ~15 min due to scheduler dedup)
5. Add .bds-cache/ to .gitignore; remove from git if previously committed
6. Push

Modes

Mode	Status
whale-radar	Shipped
token-flow	OpenClaw token-flow.mjs until ported
pulse	OpenClaw / bds-agent trade until ported
defi-analyst	Not in prefetch yet

Reference run (2026-05-27)

5 epochs, 82 trades scanned, 27 whale alerts dispatched — see operator notes in workspace daily_notes_work_plan/2026-05-27/successful-run.md.

Related

• architecture.md
• bds-api.md
• ai-coord-docs AEON_WHALE_RADAR (recipe spec)
• OpenClaw skill
• bds-agent CLI
• Powerloom Docs

License

MIT