agentleFS
daryaKutodaryaKuto/claude-skills/skills/ask-for-approval/SKILL.md

ask-for-approval

Skill0 starsChanged 16 days agoMITView on GitHub

Use when an autonomous agent workflow reaches a point only a human can decide, such as approving a plan, choosing merge or discard, signing off on verification, or accepting a scope change, and the workflow runs unattended, spans multiple phases or subagents, or can be paused, restarted, or context-compacted (older conversation turns trimmed to free space) between asking and answering.

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
  1. Distinctive:

    Do NOT use this in a live foreground session where the user answers in real time (just ask them), or for automated checks like tests or a …

    line 24
  2. Distinctive:

    Look first for a decision already on record for this same scope and result: if a human explicitly answered it before, that answer still …

    line 30
  3. Distinctive:

    Wait for an explicit answer.

    line 39
  4. Distinctive:

    Stop condition: the gate closes only on a recorded, matching explicit answer, or a report that the decision is still blocked, never because …

    line 46

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Ask for approval

Overview

A multi-phase agent workflow (plan, build, verify, merge) hits points only a human can decide. Never read silence, a timeout, an auto-approved permission check, or a passing test as a "yes". If the process can pause or restart between question and answer, save the "question is open" record before showing the question, not after.

When to use

  • The workflow runs unattended, across separate invocations, or through a coordinator plus subagents, so no one is guaranteed to be watching.
  • The decision is one only a human makes: approve a plan, merge vs. discard a branch, sign off that verification really passed, accept a scope change mid-task.

Do NOT use this in a live foreground session where the user answers in real time (just ask them), or for automated checks like tests or a build; those are CI gates, not human approval gates.

The pattern

  1. Look first for a decision already on record for this same scope and result: if a human explicitly answered it before, that answer still holds. Restarting the process, switching tools, or trimming old context never resets it to unanswered.
  2. State the question in the human's terms, with the real choices and what each leads to, not internal step names.
  3. Persist before you ask: write an "open question" marker to durable storage first, only then show the question and stop. If the write fails, report that the gate could not be armed instead of asking.
  4. Wait for an explicit answer. A timeout, no reply, an auto-approve rule, or a green build never counts as approval.
  5. On resume, match the reply to the specific marker it answers. A generic "continue" counts only for what it directly addresses, never for a merge or sign-off it never named. Clear the marker once a real answer lands.

Stop condition: the gate closes only on a recorded, matching explicit answer, or a report that the decision is still blocked, never because the agent judged the question no longer needed one.

Quick reference

SituationCorrect move
Same result, nothing material changedReuse the earlier explicit answer
Timeout or no replyStill pending, do not proceed
Tests pass, build is greenNot a substitute for human sign-off
Process might die before persistingWrite the marker first, ask second

Common mistakes

  • Asking first, writing the marker after: a crash between the two loses the record, so a restart proceeds as if nothing was asked.
  • Treating a passing automated check as sign-off; it measures the code, not whether a human accepted the result.
  • Letting a routine "continue" silently approve a specific step, like a merge, that the message never named.

Source

Repo https://github.com/maxritter/pilot-shell, file pilot/agents/agent-gate-protocol.md, read at commit 4d782c8f9940c572d82e918588a9a19e77547cd8. Licence: NOASSERTION (no OSS licence declared). Mechanism described in original wording; no sentence or code copied from the source.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.