Kbediako/evergreen-codex-skills / skills/wait-for-subagents-patiently
wait-for-subagents-patiently
VerifiedWait for quiet native Codex children and reconcile their state before declaring a stall, interrupting, replacing, or handing off without a result.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 3.4 KB |
| agents/openai.yaml | config | 239 B |
| references/budgets-and-stall-evidence.md | markdown | 3.1 KB |
| references/interruption-and-ownership.md | markdown | 1.7 KB |
| references/status-and-handoff.md | markdown | 2.1 KB |
references/budgets-and-stall-evidence.md
View on GitHub# Budgets And Stall Evidence
Elapsed time schedules work; it does not diagnose the child.
## Record A Waiting Budget
Record:
- start time;
- task complexity: small, normal, or deep;
- the next parent decision that actually needs the result;
- any user deadline, external expiry, scarce resource, or safety boundary;
- a checkpoint condition and hard decision point derived from those facts;
- the host's maximum practical blocking wait;
- whether the child blocks the parent's next action.
Do not invent generic minute tables or timer-driven status cadence. Continue independent work while useful, but a required result remains a dependency even without a deadline; use the mailbox when that result blocks the next action. Only optional work may be left off the critical path. Explicit long-running instructions, ChatGPT Pro review, browser finalization, and domain patience rules remain authoritative.
Plan checkpoints early enough for the actual deadline. At a blocking hard decision point, consume available final or checkpoint evidence; request another checkpoint and wait only when the remaining deadline, resources, and safety boundary allow it. Apply the entrypoint's urgent-stop exception when needed.
Decide whether to continue waiting, interrupt, supersede, or give an explicit partial or blocked handoff based on scope and risk. A scheduling decision neither proves a stall nor satisfies missing required acceptance evidence. Proceeding without a particular child must preserve the acceptance criteria: obtain the missing result through another authorized path or report the unmet requirement.
Leave a read-only, non-side-effecting, non-blocking child running only when the active task can safely retain it. Reconcile it before any later duplicate spawn.
## Classify Evidence
Treat these as insufficient stall evidence:
- one or more wait timeouts;
- a `running` status;
- sparse or null last-message fields;
- elapsed time alone;
- parent impatience;
- an exceeded estimate without a stated hard decision policy.
Treat these as intervention facts, not automatic stall proof:
- terminal, failed, cancelled, interrupted, or unavailable status;
- a checkpoint that reports a blocker;
- out-of-scope work, duplicate work, or unauthorized mutation;
- scarce-resource or external-side-effect consumption;
- artifact, transcript, process, log, or repo evidence that contradicts expected progress;
- no response after a hard checkpoint and expiry of a blocking hard decision point.
Call a child stalled only when at least two independent signals show lack of progress rather than quiet execution, completion, cancellation, obsolescence, or parent interruption. State the signals.
## Prefer Exact Classifications
Use:
- `quiet/running` when no update arrived but status remains active;
- `over budget` when the stated estimate or decision point elapsed;
- `obsolete` when the parent no longer needs the work;
- `timeboxed out` when the result leaves the critical path;
- `completed-on-interrupt` when the interrupt response indicates prior completion, whether or not it includes the final answer;
- `stalled` only with the evidence threshold above.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |