skills(running-in-ci): raise the bar for repo-overlay PRs#604
Merged
Conversation
Closes #603 Co-Authored-By: Claude <noreply@anthropic.com>
Merged
max-sixty
added a commit
that referenced
this pull request
May 27, 2026
## Why Cut the 0.1.2 release so consumer repos (and tend's own nightly regen) pick up the new `claude-interactive` harness and per-workflow `harness`/`model` overrides. ## What's new since 0.1.1 **New `claude-interactive` harness** (`max-sixty/tend/interactive@0.1.2`) — opt-in alternative to the released `claude` harness. PTY-supervised interactive `claude` via `script(1)`, end-of-turn detected through Stop/StopFailure hooks. Built as the trial path ahead of Anthropic's June 15 billing split between Agent-SDK metering and the flat Claude Code subscription. Smoke tested end-to-end on tend itself. PRs: #609, #611, #613, #614, #615, #616. **Per-workflow `harness` / `model` override** — adopters can flip a single workflow to a different harness or model without changing `.config/tend.yaml` defaults. #612. **Skill refinements** — nightly upstream-bot rebases (#605), running-in-ci PR bar (#604) and recheck (#573), env-filter loophole fix (#599), authorAssociation warning (#600), review-gates Gate 1 (#602). **Bug fix** — mention queue-delay now uses `comment.updated_at` so edit events report accurately (#595). ## Compatibility Released `claude` harness path is byte-identical; `claude-interactive` is strictly additive and opt-in. Consumer repos that don't touch `harness:` see no change beyond the new skill text and the mention-edit fix.
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
The bundled
running-in-ciskill'sLearning from Feedback→When to proposesection currently lets a single observed incident qualify for a repo-overlay PR (the third "Signals" bullet reads "The same correction has surfaced before, or would plausibly surface again" — the "or would plausibly surface again" clause covers any one-off failure). This produces low-signal skill churn: stochastic failures get codified as repo-specific rules that every future session then has to read past.Reported in #603. Triggering thread: PRQL/prql#5945 comment — a single ~5h40m orphan-loop incident triggered an overlay PR that was closed at maintainer request with the ask that we apply the same bar to overlay PRs that tend's own bundled-skill changes operate under.
Solution
Tighten
When to proposeso opening a repo-overlay PR requires generalizability and at least one of:The rationale: bundled-skill changes pass through human review on the tend repo, which acts as an implicit recurrence/impact filter. Per-repo overlays don't get that scrutiny, so the bar belongs in the bundled skill.
Testing
Skill text change only — no executable code touched. The rule will exercise on the next maintainer-correction in any consumer.
Closes #603 — automated triage