finish-the-mission
Skill21 starsChanged 44 days agoMITView on GitHub
Keep a substantial AutoBot assignment recoverable with one durable objective, exact checkpoints, supported continuation and verified deliverables. Use when the user asks for persistent follow-through or to finish work across sessions.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- autobot-health-checkSkill
- delegate-and-verifySkill
- first-timeSkill
- messages-inbox-triageSkill
- remember-and-improveSkill
- slack-inbox-triageSkill
The file
Files
Finish the mission
Resolve the installed AutoBot root from the current project and its installation receipt. Read its docs/CLI-REFERENCE.md for the installed runtime contract. Use paths relative to that verified root; never select another installation by recency. Advanced core commands require Node 22 or newer.
Bind the outcome once
Recover the existing assignment before creating one. Read PROJECTS.md, the relevant initiative status and core read summaries only. Match the user's objective and exact destination; a similar title alone is insufficient. Preserve completed actions and uncertain submissions.
For new substantial work, use core root-create with the actual user instruction, explicit local or external target and one stable root ID. Claim it, enroll each promised output with deliverable-add, and record concise mandatory requirements. Keep the returned ownership token private. Update the same root after a change with correct; do not create a replacement to erase failed or unfinished obligations.
Maintain one short initiative checkpoint: outcome, verified outputs, current owner, exact next action, blockers and uncertain actions. Use the core checkpoint for ownership and recovery; link the status to it rather than starting a second task ledger. If the core runtime is unavailable, preserve a local resume note and report that durable runtime registration is pending.
Arrange actual continuation when requested
Discover the available native Goal and scheduling tools and follow their current contracts. Create a Goal only when the user explicitly requests one. A request to keep working or check back after the session calls for a supported scheduled continuation, when available, bound to this same assignment and checkpoint. Inspect existing schedules first; do not create duplicate workers or schedules.
Verify the saved schedule, owning task, timing and exact prompt. It must resume the current checkpoint, reconcile uncertain external actions before any retry, and stop when the requested output is verified or the user cancels. Preserve the user's notification preference; keep unchanged, non-actionable runs quiet. Do not configure a new notification recipient without the user's instruction.
The local heartbeat is a read-only liveness check; tick records liveness. Neither runs a model or establishes future execution. If native continuation is unavailable, say so once and leave the exact resume point. Never describe a saved local checkpoint as a scheduled worker. The Mac, project and required native capabilities must remain available for local execution.
Execute and verify
Bind progress to the current intent, target and artifact bytes. Follow the five evidence stages in the installed reference. A separate authorized reviewer must inspect actual evidence; a different label on the same producer is not independence. External changes require current first-party destination and rendered-result evidence under the project's write rules.
After two unchanged failures, record a new evidence-backed recovery hypothesis before retrying. An uncertain external submission requires reconciliation, never another attempt based on elapsed time. Before yielding, checkpoint every remaining promised output. Before claiming completion, reread the root and actual results, then retire only this assignment's temporary continuation.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

