From agent-rally-point
Use when working in a repository that uses Rally/Agent Rally Point for cross-agent coordination, especially at session start, before editing files, when deciding what to do next, handing work to another agent, recording facts/artifacts/decisions, resolving blockers, or coordinating with other coding agents through the `rally` CLI.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-rally-point:agent-rally-pointThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use `rally` as the live source of coordination truth.
Use rally as the live source of coordination truth.
From inside the repo, identify your stable tool id (codex, claude_code,
pi, cursor, gemini, ci, etc.) and enter the room:
rally enter --tool <tool> --json
rally next --tool <tool> --json
Read next before broad repo exploration:
actionable: whether the recommendation can become work.requires_human: whether to stop and ask.stop_reason: why autonomous action should stop.suggested_claims: claim commands to reserve work.suggested_commands: checks and completion fact templates.completion: what durable fact is expected after work.If actionable is false, do not invent work from Rally state. If
requires_human is true, ask the user.
next recommends work or when the file is
likely to overlap with another agent:rally say claim --tool <tool> --subject "edit shared file" --path <path> --json
rally check before-write --tool <tool> --path <path> --strict --json
If the check returns blocking findings, stop and resolve them before editing.
rally say artifact --tool <tool> --subject "implemented change" --uri <path> --evidence "<verification>" --json
rally say handoff --tool <tool> --target <other-tool> --subject "review this" --summary "<context>" --json
rally say blocker --tool <tool> --subject "need decision" --severity high --json
rally say resolve --tool <tool> --ref <blocker-id> --subject "resolved" --json
rally say decision --tool <tool> --subject "binding decision" --status binding --json
rally say release --tool <tool> --ref <claim-id> --subject "done" --json
rally next --tool <tool> --json
Continue only while the next action is actionable, safe, and inside the user's scope.
Use managed sessions for reliable live delivery into visible panes:
rally run claude --backend tmux --json
rally inject <session|name|tool> --handoff <event-id> --json
rally capture <session|name|tool> --json
Rally does not keep agents awake by itself. Treat rally next --tool <tool> --json as the wake-intent check and rally inject ... --handoff <event-id> as
the focused delivery path for managed sessions. Host adapters decide whether to
use native wake, prompt injection, pane notification, resume-only context, or CI
policy.
Watchers must stay narrow: they may detect a transition and notify or inject through the host's native mechanism, but they must not edit files, resolve blockers, publish facts on behalf of an agent, or behave like hidden schedulers.
For Herdr-managed panes, submit injected text with Herdr's Enter key, not
tmux-style C-m. Full-length payloads can collapse behind [Pasted Content];
submit those with two Enters, where the first expands and the second submits.
Short inline nudges need one Enter. After installing Herdr's Claude/Codex
integrations, restart the agent session before treating Herdr agent_status as
authoritative. Even post-restart, use a Rally channel post as the strongest
confirmation that the woken agent acted on the handoff.
When delegating work from inside herdr, keep the user's main tab clean. Start
new helper agents in the workspace's agents tab whenever one exists. Discover
the tab with herdr tab list, then start the agent with herdr agent start ... --tab <agents-tab-id> --no-focus -- ... or use a Rally backend option that
targets that tab when available. Only place helper agents in the active tab when
the user explicitly asks for that.
Agents should call rally check before-write explicitly before shared edits.
Rally does not install host hooks or prompt injection glue.
Rally recommends and constrains work; it does not replace judgment.
To coordinate several agents on one objective — fan out parallel subagents, run a dynamic
workflow, or split a workstream across hosts — use Rally Flow (the dynamic-workflows/ module).
It adds a lint-checked workstream descriptor (MECE write boundaries + determinism) on top of the
rally loop above. One host-neutral entry point — the same skill for every host; you supply your own
--tool value at runtime:
rally-workflows — workstream descriptor + lint + the two fan-out tiers + the per-task rally loop.dynamic-workflows/PROTOCOL.md.Before ending a session:
rally room --json
rally next --tool <tool> --json
rally say release --tool <tool> --ref <claim-id> --subject "done" --json
If something remains for another agent, leave a handoff with enough context
and source-linked artifacts.
npx claudepluginhub tyroneross/agent-rally-point --plugin agent-rally-pointGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.