docs(build-agent): document harness.permissions.default_mode for testing gated writes#17
Open
bekossy wants to merge 1 commit into
Open
docs(build-agent): document harness.permissions.default_mode for testing gated writes#17bekossy wants to merge 1 commit into
bekossy wants to merge 1 commit into
Conversation
…ing gated writes Hit a real stall while testing a multi-tool GitHub+Slack agent: SEND_MESSAGE tripped an approval gate through test-agent.sh's low-level invoke, which doesn't hydrate a resumable Playground session, so the pending approval had nothing to resolve against and never completed. The identical config ran clean through a real Playground chat session instead. harness.permissions.default_mode: "bypassPermissions" (the same field behind the Playground's Advanced panel "Bypass (skip every gate)" option, per web/packages/agenta-entity-ui/.../ClaudePermissionsControl.tsx in the agenta repo) skips approval gates entirely. Verified end to end: the same previously-stalling SEND_MESSAGE call completed with an ok result and a PASS check-tools.sh verdict once this was set. Documented as an explicit testing convenience, not added to the fixed boilerplate - it's a deliberate opt-in for whoever is testing, not something that should ship silently in every agent config.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Context
Per today's standup (2026-07-08), the team flagged "approval-related agent breaks during testing" as something to fix by configuring bypass-permission auto-approval. Hit exactly this while testing a multi-tool GitHub + Slack agent:
SEND_MESSAGEtripped a human-approval gate throughtest-agent.sh's low-level invoke. That endpoint doesn't hydrate a resumable Playground session, so the pending approval had nothing to resolve against and stayed stuck (APPROVAL: paused,check-tools.shverdictUNCONFIRMED). The identical config ran clean through a real Playground chat session instead, since that session support actually resolving the approval.Changes
Documented
harness.permissions.default_mode: "bypassPermissions"inSKILL.md, right where the approval-gate behavior is already explained. It's the same field behind the Playground's Advanced panel "Bypass (skip every gate)" option (confirmed inweb/packages/agenta-entity-ui/src/DrillInView/SchemaControls/ClaudePermissionsControl.tsxin theagentarepo).Explicitly kept out of the fixed boilerplate. It skips every approval gate for the whole run, including ones a real end user would want to review, so it's documented as a deliberate opt-in for whoever is testing, not a default.
Tests
Verified end to end: added this field to the exact multi-tool config that previously stalled, re-ran through
test-agent.sh. Before:APPROVAL: paused on SEND_MESSAGE,check-tools.shverdictUNCONFIRMED. After:SEND_MESSAGEcompleted with anokresult inTOOLS, noAPPROVALline,check-tools.shverdictPASS. Confirmed the message actually posted by reading#non-human-chatback.What to QA
SEND_MESSAGE), test it throughtest-agent.shwithout this field set. It should stall onAPPROVAL: pausedwith anUNCONFIRMEDcheck-tools.shverdict.harness.permissions.default_mode: "bypassPermissions", commit the change, re-test. The write should complete with anokresult and aPASSverdict.