fix(ulw-plan): recognize approval intent and stop the approval-gate loop#54
Closed
JSap0914 wants to merge 1 commit into
Closed
fix(ulw-plan): recognize approval intent and stop the approval-gate loop#54JSap0914 wants to merge 1 commit into
JSap0914 wants to merge 1 commit into
Conversation
The planner waited for one exact approval phrase before writing `.omo/plans/<slug>.md` and kept no durable record of what it was waiting on, so after context growth it re-ran exploration and re-emitted the full blocked-gate message — a token-burning loop even when the user clearly approved. Tighten the gate wording so the planner accepts common approval replies (and localized equivalents), persists the pending gate to `.omo/drafts/<slug>.md` and reads that state on the next turn instead of restating the whole plan, and emits a single concise prompt when still unapproved. The planner-only safety boundary and the `$start-work` bootstrap exception are unchanged. Extends ulw-plan-skill.test.mjs. Fixes code-yeongyu#48.
Contributor
|
Thanks for the PR! Changes to LazyCodex land through oh-my-openagent — could you open this against omo-codex over there instead? PRs in this repository can't be merged. If you're working with a coding agent, prompt it like this:
|
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
Updates the ulw-plan skill documentation and tests to make the “approval gate” more robust and less prone to looping by persisting gate state and accepting flexible approval intent.
Changes:
- Added new “recognize approval intent / persist gate / avoid looping” guidance to
SKILL.mdandfull-workflow.md. - Mirrored the same approval-gate guidance into the
components/ultraworkcopies of the docs. - Strengthened tests to assert the new guidance is present in the skill/workflow references.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| plugins/omo/test/ulw-plan-skill.test.mjs | Adds assertions to enforce new approval-gate wording in docs. |
| plugins/omo/skills/ulw-plan/references/full-workflow.md | Documents robust approval handling + durable gate state persistence. |
| plugins/omo/skills/ulw-plan/SKILL.md | Adds a compact bullet describing the new approval gate behavior. |
| plugins/omo/components/ultrawork/skills/ulw-plan/references/full-workflow.md | Mirrors the same workflow guidance for the ultrawork component. |
| plugins/omo/components/ultrawork/skills/ulw-plan/SKILL.md | Mirrors the same SKILL.md bullet for the ultrawork component. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+31
to
+34
| assert.match(skill, /Recognize approval intent/); | ||
| assert.match(skill, /do not demand one exact magic phrase/); | ||
| assert.match(skill, /record the pending gate in `\.omo\/drafts\/<slug>\.md`/); | ||
| assert.match(skill, /ask ONE concise approval question/); |
|
|
||
| Then **wait for the user's explicit okay** before generating the plan. No Metis, no plan file, no execution until the user approves. If the user amends scope, fold it in and re-present the brief. This gate replaces any automatic interview-to-plan transition. | ||
|
|
||
| Handle approval intent robustly so the gate never loops. Accept clear approval replies — "yes", "approve", "approved", "proceed", "go ahead", "do it", "write the plan", "create the plan", and localized equivalents (e.g. Korean "네", "응", "진행", "ㄱㄱ", "작성해") — as approval; never require one exact magic phrase. When you present the brief, persist the pending gate to `.omo/drafts/<slug>.md` with the exact pending action ("awaiting approval to write `.omo/plans/<slug>.md`") and target plan path; on the next turn read that durable gate state first instead of re-running exploration or re-explaining the full planning status. If the reply is still not approval, emit a single concise approval prompt rather than restating the whole brief, so the gate cannot burn context in a re-approval loop. Replies that amend scope update the brief and re-present it; they never trigger plan generation. |
|
|
||
| Then **wait for the user's explicit okay** before generating the plan. No Metis, no plan file, no execution until the user approves. If the user amends scope, fold it in and re-present the brief. This gate replaces any automatic interview-to-plan transition. | ||
|
|
||
| Handle approval intent robustly so the gate never loops. Accept clear approval replies — "yes", "approve", "approved", "proceed", "go ahead", "do it", "write the plan", "create the plan", and localized equivalents (e.g. Korean "네", "응", "진행", "ㄱㄱ", "작성해") — as approval; never require one exact magic phrase. When you present the brief, persist the pending gate to `.omo/drafts/<slug>.md` with the exact pending action ("awaiting approval to write `.omo/plans/<slug>.md`") and target plan path; on the next turn read that durable gate state first instead of re-running exploration or re-explaining the full planning status. If the reply is still not approval, emit a single concise approval prompt rather than restating the whole brief, so the gate cannot burn context in a re-approval loop. Replies that amend scope update the brief and re-present it; they never trigger plan generation. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #48.
Problem
ulw-plan's approval gate waits for the user to reply with one exact phrase before it writes.omo/plans/<slug>.md, and it keeps no durable record of what it is waiting on. After context growth or a turn transition the planner re-runs exploration and re-emits the full "blocked at the approval gate" message — a token-burning loop, even when the user has clearly approved ("yes", "proceed", etc.).Fix
Tighten the gate wording in the
ulw-planskill and itsreferences/full-workflow.mdso the planner:.omo/drafts/<slug>.md(exact pending action + target plan path) and reads that durable state on the next turn instead of re-running exploration or restating the whole plan;The safety boundary is unchanged: still planner-only, still no plan file / Metis / execution before approval, and the narrow
$start-workbootstrap exception is preserved.Verification
Extended
plugins/omo/test/ulw-plan-skill.test.mjswith assertions for the approval-intent / durable-state / single-prompt contract:Component (
components/ultrawork/skills/ulw-plan/) and packaged (skills/ulw-plan/) copies are kept byte-identical, so the sync-skills drift check stays green.Summary by cubic
Stops the approval-gate loop in
ulw-planby recognizing approval intent and persisting gate state to.omo/drafts/<slug>.md. Fixes #48 while keeping the planner-only safety boundary and the$start-workexception unchanged..omo/plans/<slug>.md) and read it on the next turn to avoid re-running exploration.Written for commit e3c3f10. Summary will update on new commits.