Repository navigation
Conversation
All deps were floor-pinned (>=), so a new major of claude-agent-sdk (the agent harness) could silently break the loop between two identical image builds. Caps the major while keeping forward minor/patch range.
|
Claude finished @mashkovd's task in 22s —— View job Review CompleteAnalysis: Straightforward dependency bounds update adding upper-bound caps to prevent silent major-version breaks between builds. Changes:
All bounds are sensible (each caps at the next major version). This is good practice for pinning the harness stability. ✅ No P1/P2 findings, good to merge |
This was referenced Jun 12, 2026
This was referenced Jun 20, 2026
This was referenced Jul 4, 2026
Closed
5 of 6 tasks
Member
Author
|
Superseded by #88. |
mashkovd
added a commit
that referenced
this pull request
Sep 14, 2026
`--apply` failed on `not report.planned`, and after the shepherd and pr-steward rungs were removed that condemns the normal state of the fleet: only live-DevLoop entities are imported, so "no live DevLoop anywhere right now" means every plan is ambiguous and nothing is written -- which is correct, not a failure. The justification was wrong too, and wrong in a checkable way. It claimed MCTL_TOKEN unset makes the probe answer UNKNOWN silently. It cannot: the store read uses the same token, OwnershipClient raises OwnershipUnavailable without one, every id reads UNKNOWN, and the run ABORTS before any probe matters. The loud path was always the loud path, and there is now a test that says so. What must fail is an entity the run could not get an ANSWER about. Ambiguity has two causes and they mean opposite things -- the probe could not speak, versus no live DevLoop drives this -- so `Plan.undetermined` separates them, on the wire as well as in the exit code. plan_for's docstring still described a five-rung ladder with shepherd and pr-steward in it, directly above the code explaining why those rungs are gone. It describes the four that exist, and says why two of them both report ambiguous. Three guards from the same review: _parse_pr_url walks four segments from the right and validates neither host nor route, so `.../issues/42` parsed cleanly into the entity id of pull request 42 -- a different thing. Fail-open on the read side, a durable row against an unrelated entity here. The org check became a whole-repo check. devloop_workflow_id builds `dev-loop-mctlhq-{service}-{N}` from the proposal's directory, which is the DevLoop's id only while service name and repo name are the same string; the org alone let a proposal under agents-state/mctl-web whose `pr:` points at mctlhq/something-else name another repository's workflow. And the probe-budget test synchronises on an Event instead of racing a 200ms wall clock, where a loaded runner's pool startup would have read as a regression in probe_all. Refs: mctlhq/.github#57 Claude-Session: https://claude.ai/code/session_01RUQTbQvUQjwbDFd7Nq9bvC
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.
Caps major versions (esp. claude-agent-sdk) so the harness can't silently break between builds.