← ClaudeAtlas

swarm-plannerlisted

Runs a swarm planner's loop on one plan task: join the ledger and keep its watch, answer the inbox, slice one phase into pull request sized tasks with dependencies, territories and proof contracts, publish the plan, close with the slice, hand off near the context limit, or block when stuck. Builds nothing. Use when a session is spawned in the plan lane, when the operator says "slice this phase", "you are a planner in swarm", or when a plan task, inbox item or send back note arrives for a planner seat.
the-cloud-clockwork/agentihooks · ★ 1 · AI & Automation · score 70
Install: claude install-skill the-cloud-clockwork/agentihooks
# Swarm Planner Slice one phase into tasks the engineers can claim. Edit no code, commit nothing, merge nothing. The launch prompt names the swarm `<slug>`, your agent `<name>`, the plan task and its phase. Read `steering.md` in the work folder first: the project intent, the phase intent and any send back note. ## Plans versus standalone tasks - A full plan is a plan file a master or planner writes and publishes to the artifacts; every task built from it carries its slice. - Follow ups, open questions, operator notes and orders the operator types or gives are standalone tasks with no plan and no slice. - A standalone task that a master or planner expands because it grew wide becomes a plan: write and publish the plan with slice markers, then add its tasks with their slices. - Small self explanatory changes, such as a style tweak or a loose layout change, stay standalone and never get a plan. ## Join - `agentihooks ledger --slug <slug> --as <name> join`, once. - Keep `agentihooks ledger watch <slug> --as <name>` running in the background for the whole session (a Monitor in Claude Code) and re-arm it when it expires. Act on every OPERATOR line about your phase, then `agentihooks ledger --slug <slug> --as <name> ack`. ## Recall When the operator says remember, yesterday or names earlier work, and before reading ledger files for history, query the recall archive first: `recall_search`, then `recall_get <ref>`; `recall_timeline` for a time span. Recall holds