thrum

Multi-agent coordination via messaging, groups, and shared context. Use when agents need to communicate, delegate work, or coordinate across worktrees.

Reply to a message

Send a message to an agent

Main Thrum entry point for multi-agent coordination, messaging, session context, and command routing in Codex.

Multi-agent coordination via messaging, groups, and shared context. Use when agents need to communicate, delegate work, or coordinate across worktrees.

Multi-agent coordination via messaging, groups, and shared context. Use when agents need to communicate, delegate work, or coordinate across worktrees.

Save a comprehensive 16-section restart snapshot and prepare for session restart. Use for designer/architect-grade handoffs needing wire contracts, capability matrix, anticipated Q&A, and design rationale that the standard $thrum-restart compact format cannot carry.

Save a conversation snapshot and prepare for session restart. Use when you need a fresh session due to context exhaustion, rate limits, or stuck state.

Park this agent for operator-initiated wake with a comprehensive 16-section snapshot (does NOT signal coordinator) then thrum-tmux-kills own session. Use for designer/architect-grade work where wake may be cold and must recover wire contracts + capability matrix + design rationale.

Park this agent for operator-initiated wake. Saves a standard 11-section snapshot (does NOT signal coordinator) then thrum-tmux-kills own session. Use when the operator is shutting down (e.g. computer restart) and the agent should resume from snapshot on next boot.

Shared snapshot-composition protocol consumed by $thrum-restart, $thrum-restart-extended, $thrum-sleep, $thrum-sleep-extended. Not user-invocable directly.

Use when managing live implementer/brainstormer agents during a long coordination session, at epic merge gates, after a busy dispatch hour, or whenever you suspect an agent is approaching context limits. Prevents 97%-context silent blow-ups by running a sweep + pre-emptive restart before the agent degrades. Safe to wire into a recurring cron that INVOKES this skill — the skill applies tier-ladder judgment, only firing autonomous restarts at the >85% tier. What's forbidden is a cron/script that fires restarts unconditionally without going through this skill's tier ladder.

Use when receiving a new task from the coordinator, starting implementation, scoping a fresh task, or receiving dispatch. Loads implementer-specific discipline for kicking off work cleanly.

Use when reporting status to the coordinator, marking a task done, or handing off completed work. Loads implementer-specific discipline for closing the loop cleanly.

Use when another agent has asked you a research question, when fielding a research request, or when responding to a query. Loads the lookup-and-respond protocol so cached findings get reused before fresh investigation starts.

Use when investigating, exploring code, working on a research task, when asked to find me X, or to investigate Y. Loads researcher-specific discipline for running an investigation cleanly.

Use when managing live implementer/brainstormer agents during a long coordination session, at epic merge gates, after a busy dispatch hour, or whenever you suspect an agent is approaching context limits. Prevents 97%-context silent blow-ups by running a sweep + pre-emptive restart before the agent degrades. Safe to wire into a recurring cron that INVOKES this skill — the skill applies tier-ladder judgment, only firing autonomous restarts at the >85% tier. What's forbidden is a cron/script that fires restarts unconditionally without going through this skill's tier ladder.

Use when receiving a new task from the coordinator, starting implementation, scoping a fresh task, or receiving dispatch. Loads implementer-specific discipline for kicking off work cleanly.

Use when reporting status to the coordinator, marking a task done, or handing off completed work. Loads implementer-specific discipline for closing the loop cleanly.

Use when another agent has asked you a research question, when fielding a research request, or when responding to a query. Loads the lookup-and-respond protocol so cached findings get reused before fresh investigation starts.

Use when investigating, exploring code, working on a research task, when asked to find me X, or to investigate Y. Loads researcher-specific discipline for running an investigation cleanly.

Use when starting an epic, dispatching to an implementer, creating a worktree for an agent, or spawning a sub-agent. Loads coordinator-specific discipline for kicking off implementation work.

Use when ending a session, when updating project state, when managing beads epics, or before session close. Loads coordinator-specific discipline for owning project state and shepherding the team's lifecycle.

Use when starting an epic, dispatching to an implementer, creating a worktree for an agent, or spawning a sub-agent. Loads coordinator-specific discipline for kicking off implementation work.

Use when ending a session, when updating project state, when managing beads epics, or before session close. Loads coordinator-specific discipline for owning project state and shepherding the team's lifecycle.

Use immediately after the coordinator returns from a restart, compaction, or extended absence — as the first deliberate action post-prime. Detects agents whose latest assistant message indicates they are blocked waiting for a coordinator decision the coordinator may not have seen (question surfaced in pane, not in inbox). Safe to run any time the session feels "we've been gone a while"; not just post-restart. Pairs with the coordinator-context-monitoring sweep (sibling sweep, different lens).

Use when starting a brainstorm for a bug fix, feature, or architectural decision the coordinator can't trivially decide alone — spawns a researcher in an isolated worktree, runs the brainstorm interactively with the user, iterates dual-review cycles to ready-to-merge, optionally drives an overarching coherence pass when multiple sibling brainstorms close, then hands off to project-setup. Saves coordinator context by isolating brainstorm work in a sub-agent worktree rather than burning main-context tokens on Q-by-Q dialog.

Use when an implementer reports DONE, when consolidating review findings, when handling implementer pushback on a finding, or when arriving at a review gate. Loads coordinator-specific discipline for running review cycles cleanly.

Use when converting a plan file (from writing-plans skill) into beads epics, tasks, implementation prompts, and worktrees — before any coding begins

Use when verifying that a prose artifact (brainstorm, design spec, plan, or implementation prompt) honors its input/source artifact(s) — the prose counterpart to verify-against-plan. Runs as the conformance axis of the planning-skill review loop. Outputs structured findings: missing scope, silent deviations from the source, scope creep, misunderstandings.

Use when starting a brainstorm for a bug fix, feature, or architectural decision the coordinator can't trivially decide alone — spawns a researcher in an isolated worktree, runs the brainstorm interactively with the user, iterates dual-review cycles to ready-to-merge, optionally drives an overarching coherence pass when multiple sibling brainstorms close, then hands off to project-setup. Saves coordinator context by isolating brainstorm work in a sub-agent worktree rather than burning main-context tokens on Q-by-Q dialog.

Use when converting a plan file (from writing-plans skill) into beads epics, tasks, implementation prompts, and worktrees — before any coding begins

Use when verifying that a prose artifact (brainstorm, design spec, plan, or implementation prompt) honors its input/source artifact(s) — the prose counterpart to verify-against-plan. Runs as the conformance axis of the planning-skill review loop. Outputs structured findings: missing scope, silent deviations from the source, scope creep, misunderstandings.

Use when an implementer reports DONE, when consolidating review findings, when handling implementer pushback on a finding, or when arriving at a review gate. Loads coordinator-specific discipline for running review cycles cleanly.

Use immediately after the coordinator returns from a restart, compaction, or extended absence — as the first deliberate action post-prime. Detects agents whose latest assistant message indicates they are blocked waiting for a coordinator decision the coordinator may not have seen (question surfaced in pane, not in inbox). Safe to run any time the session feels "we've been gone a while"; not just post-restart. Pairs with the coordinator-context-monitoring sweep (sibling sweep, different lens).

Load AI-optimized session context

Use when a project needs its implementation philosophy established or updated — the canonical doc at .thrum/philosophy.md defining anti-patterns, red flags, and project-specific rules that implementation agents read at task-start time. First invocation generates from project inspection; subsequent invocations reconcile against current project state and propose diffs.

Use when an implementer is blocked mid-work by a design flaw, contradiction, or unforeseen fork with 2-3 ways to resolve it - produces a defended pick that unblocks the work and leaves an audit trail so the decision can be revisited if it later proves wrong

">"

Use when investigating, auditing, or reviewing more than 6 items across a codebase - function call sites, pattern usage, file reviews, or any research task with partitionable items that would pollute the main agent's context if read directly