agentsclimarketplace

Agent rally watcher

Skill tyroneross/build-loop/skills/agent-rally-watcher

Use when listening for Rally Point changes, wiring coordination watchers, debugging watch-loop behavior, or changing the future agent-rally-watcher spin-out surface.From its SKILL.md

Install
npx -y skills add tyroneross/build-loop --skill agent-rally-watcher

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.1 KB, as published. Nobody here has run it

<!-- SPDX-FileCopyrightText: 2025-2026 Tyrone Ross, Jr <[email protected]> | SPDX-License-Identifier: Apache-2.0 -->

Agent Rally Watcher

Build-loop ships a native embedded agent-rally-watcher capability while the standalone watcher repo matures. Treat it as a mini-plugin inside build-loop: watcher behavior lives in one namespace, and build-loop keeps compatibility entrypoints thin.

Native Surface

PurposePath
Watcher package${CLAUDE_PLUGIN_ROOT}/scripts/agent_rally_watcher/
Compatibility CLI${CLAUDE_PLUGIN_ROOT}/scripts/coordination_watch.py
Status dependency${CLAUDE_PLUGIN_ROOT}/scripts/coordination_status.py
Manual command${CLAUDE_PLUGIN_ROOT}/commands/rally-point.md watch subcommand
Boundary manifest${CLAUDE_PLUGIN_ROOT}/scripts/rally_point/plugin_boundary.json

Operating Rules

  • Keep watch-loop logic under scripts/agent_rally_watcher/.
  • Keep scripts/coordination_watch.py as a compatibility wrapper.
  • Watchers emit compact JSONL transition events; they do not make decisions, mutate code, stage files, commit, or resolve verdicts.
  • Prefer short interactive watchers for active work. Durable always-on watchers belong in launchd/systemd packaging, not ad hoc shell backgrounding.
  • Stop watchers started by the current run before final closeout.

Commands

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/coordination_watch.py \
  --workdir "$PWD" \
  --session-id "$SESSION_ID" \
  --tool "$TOOL_ID" \
  --task-ref "$TASK_REF" \
  --interval 5 \
  --jsonl \
  --baseline-current

For one-shot validation:

python3 ${CLAUDE_PLUGIN_ROOT}/scripts/coordination_watch.py \
  --workdir "$PWD" \
  --session-id watcher-smoke \
  --tool codex \
  --iterations 1 \
  --jsonl

Watcher transition events include inbox_latest_messages, a compact doorbell preview of the newest addressed/broadcast inbox records. It is additive to the raw unread counts; read the inbox before acting on the full message.

Watcher transition events also include task_heartbeat. Pass --task-ref when a long-running task has an expected claim/run/pillar id; the watcher will wake on current/stale/wrong-task/blocked heartbeat changes without requiring the peer to send an inbox message.

Validation

uv run pytest scripts/test_coordination_status.py scripts/rally_point/test_session_probe.py
python3 scripts/coordination_watch.py --workdir "$PWD" --session-id watcher-smoke --tool codex --iterations 1 --jsonl
python3 scripts/agent_rally.py boundary --repo "$PWD" --check --json

Spin-Out Rule

When extracting to the standalone watcher plugin, copy scripts/agent_rally_watcher/, this skill, watcher tests, and the compatibility contract named in plugin_boundary.json. Build-loop should then keep coordination_watch.py as a thin adapter to the standalone package.

Keep looking

Skills are one crate of 608,666. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.