agent-state-machine
Skill11 starsChanged 3 months agoMITView on GitHub
Use when designing or implementing states for AI agent tasks, background jobs, local model runs, queues, and notifications — one shared set of states, only legal transitions, and a recovery path out of every failure.
In this repository39 other files it gives its agents.
- agent-starter-packSkill
- local-ai-ollamaSkill
- apple-hig-command-paletteSkill
- apple-hig-empty-statesSkill
- apple-hig-feedback-statusSkill
- apple-hig-inspectorsSkill
- apple-hig-macos-window-layoutSkill
- apple-hig-search-filteringSkill
- apple-hig-settingsSkill
- apple-hig-sf-symbolsSkill
- apple-hig-sidebarsSkill
- apple-hig-toolbarsSkill
- apple-hig-typographySkill
- blog-post-publishingSkill
- social-post-kitSkill
- color-principlesSkill
- kika-design-systemSkill
- emergency-switch-appSkill
- product-hunt-polish-reviewSkill
- docs-updateSkill
- project-folder-cleanupSkill
- between-runs-auditSkill
- explain-new-projectSkill
- new-project-gateSkill
- old-project-auditSkill
- project-catch-upSkill
- project-workflowSkill
- project-worth-my-timeSkill
- butterfly-effectSkill
- spreadSkill
- tilSkill
- apple-app-intentsSkill
- macos-app-distribution-dmgSkill
- macos-clipboard-pasteboardSkill
- macos-global-shortcutsSkill
- macos-launch-at-loginSkill
- macos-menubar-swiftuiSkill
- macos-notificationsSkill
- macos-permissions-privacySkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 80
Avoid impossible transitions unless explicitly handled.
- Distinctive: line 202
Do not notify for every transition.
- Distinctive: line 206
Use real progress only when measurable.
- Distinctive: line 242
Every failed / blocked / waiting state needs a way out.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent State Machine
Model every agent / task / run with one shared state machine: a fixed set of states, only the transitions that are legal between them, and a recovery path out of every failure or wait. This kills the most common agent-UX bug — random status strings invented per screen.
When to use
- Active runs, task queues, AI generation, local model jobs
- Agent progress, notifications, error recovery
- Retry / stop / cancel behavior and "needs review" states
Core rule
Agent UX needs clear states, legal transitions, and recovery paths. Do not invent random labels per screen.
State vocabulary
Draft
Queued
Starting
Thinking
Running
Waiting
Blocked
Paused
Cancelling
Cancelled
Failed
Completed
Needs Review
ArchivedState meanings
Draft Created but not ready to run
Queued Ready and waiting
Starting Preparing environment/model/files
Thinking Planning/reasoning before action
Running Executing work
Waiting Waiting for external service/user/system
Blocked Cannot continue without user action
Paused Temporarily stopped, can resume
Cancelling Stop requested, cleanup in progress
Cancelled Stopped by user/system
Failed Ended with error
Completed Finished successfully
Needs Review User should inspect result
Archived Hidden from active workflowLegal transitions
Draft → Queued
Queued → Starting
Starting → Thinking
Thinking → Running
Running → Waiting
Waiting → Running
Running → Needs Review
Running → Completed
Running → Failed
Running → Blocked
Running → Cancelling
Cancelling → Cancelled
Failed → Queued
Blocked → Queued
Needs Review → Archived
Completed → Archived
Paused → RunningAvoid impossible transitions unless explicitly handled.
Reference implementation
Swift shown below, but the pattern is language-agnostic — one enum, one validator.
enum AgentRunState: String, Codable, CaseIterable, Hashable {
case draft
case queued
case starting
case thinking
case running
case waiting
case blocked
case paused
case cancelling
case cancelled
case failed
case completed
case needsReview
case archived
var title: String {
switch self {
case .draft: "Draft"
case .queued: "Queued"
case .starting: "Starting"
case .thinking: "Thinking"
case .running: "Running"
case .waiting: "Waiting"
case .blocked: "Blocked"
case .paused: "Paused"
case .cancelling: "Cancelling"
case .cancelled: "Cancelled"
case .failed: "Failed"
case .completed: "Completed"
case .needsReview: "Needs Review"
case .archived: "Archived"
}
}
var symbol: String {
switch self {
case .draft: "doc"
case .queued: "list.bullet.rectangle"
case .starting: "play.circle"
case .thinking: "brain"
case .running: "bolt.circle"
case .waiting: "clock"
case .blocked: "xmark.octagon"
case .paused: "pause.circle"
case .cancelling: "xmark.circle"
case .cancelled: "xmark.circle"
case .failed: "exclamationmark.triangle"
case .completed: "checkmark.circle"
case .needsReview: "checkmark.seal"
case .archived: "archivebox"
}
}
}State transition validator:
extension AgentRunState {
func canTransition(to next: AgentRunState) -> Bool {
switch (self, next) {
case (.draft, .queued),
(.queued, .starting),
(.starting, .thinking),
(.thinking, .running),
(.running, .waiting),
(.waiting, .running),
(.running, .needsReview),
(.running, .completed),
(.running, .failed),
(.running, .blocked),
(.running, .cancelling),
(.cancelling, .cancelled),
(.failed, .queued),
(.blocked, .queued),
(.needsReview, .archived),
(.completed, .archived),
(.paused, .running):
return true
default:
return false
}
}
}User-facing actions per state
Draft: Edit, Queue, Delete
Queued: Run Now, Remove from Queue, Edit
Starting: Cancel
Thinking: Cancel
Running: Stop, Open Logs
Waiting: Retry, Cancel, Open Settings if service issue
Blocked: Resolve, Retry, Cancel
Paused: Resume, Cancel
Failed: Retry, View Error, Archive
Completed: Review, Archive, Run Again
Needs Review: Review, Approve, Archive
Archived: RestoreNotifications per state
Notify only for:
Completed
Failed
Blocked
Needs ReviewDo not notify for every transition.
Progress rules
Use real progress only when measurable. For agent tasks, prefer step labels over a fake bar:
Planning
Editing files
Running checks
Summarizing changes
Waiting for reviewAvoid fake percentages.
Persistence fields
Each run should store:
id
title
projectId
state
createdAt
startedAt
updatedAt
completedAt
model
provider
currentStep
errorMessage
outputSummary
outputPath
retryCountRecovery rules
Every failed / blocked / waiting state needs a way out.
Failed → show error + Retry
Blocked → show what the user must fix
Waiting → show what service/input is missing
Offline → open settings / retryReview checklist
[ ] Uses one shared state enum
[ ] States have clear user-facing titles
[ ] States have consistent icons
[ ] Illegal transitions are prevented or handled
[ ] Failed/blocked/waiting states have recovery
[ ] Notifications fire only for important states
[ ] No fake progress percentages
[ ] Needs Review exists before archive/done
[ ] Retry behavior is defined
[ ] Cancel/stop behavior is definedCommon mistakes
Random status strings across files
Done/failed only, no waiting/blocked/review
No transition rules
No recovery actions
Notification spam
Fake progress
Cancelling instantly becomes cancelled without cleanup
Completed work disappears before reviewPrompt template
Drop this into any coding agent (Claude Code, Codex, Cursor, …) to apply the skill:
Use the agent-state-machine skill to implement or clean up agent/task/run states.
Rules:
- Use a single shared state enum.
- Use clear user-facing titles and consistent icons.
- Define legal transitions and prevent or handle illegal ones.
- Add user-facing actions per state.
- Add recovery for failed, blocked, waiting, and offline states.
- Notify only for completed, failed, blocked, and needs-review.
- Avoid fake progress percentages.
- Store enough run metadata for review and retry.
After coding:
1. List states added/changed.
2. Explain transition rules.
3. Explain actions per state.
4. Explain notification behavior.
5. Give manual test steps.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.

