Skip to content

Write down what the first real run found about its own runner - #133

Merged
DeDuva merged 1 commit into
devfrom
chore/dependabot-prompt-env-fixes
Sep 1, 2026
Merged

Write down what the first real run found about its own runner#133
DeDuva merged 1 commit into
devfrom
chore/dependabot-prompt-env-fixes

Conversation

@DeDuva

@DeDuva DeDuva commented Sep 1, 2026

Copy link
Copy Markdown
Owner

Follow-up to #132, from the first MODE=merge run of the routine (session, 62 turns, 888s). That run merged #131, #129 and #90 and commented on all 14 open Dependabot PRs, so the rules held up — but it also exposed five things the prompt got wrong about the environment it actually runs in.

# Found Fix
1 gh is not installed in the cloud runner Say so, and give the translation: MCP tools for mutations, curl for bulk reads
2 /advisories is blocked"sessions are bound to their configured repositories" Use npm audit --package-lock-only for the security signal
3 npm is 10.9.7, not 11 Scope the allow-scripts note to npm 11+
4 MCP appends an AI attribution trailer to every comment Read the comment back and edit it off; CLAUDE.md forbids it
5 Auto-merge refuses an already-green PR Document that this is the expected answer, and merge directly

On (1)

The run recovered on its own — it probed for gh, found none, and switched to mcp__github__* plus unauthenticated REST. That worked, but nothing in the document told it to, and a run that improvises its way around a missing tool can equally improvise its way around a safety rule. The gh commands stay, since they are the clearest statement of what to fetch; what is new is a table saying which transport to use for what, and that the curl path is read-only and repo-scoped.

On (4)

Not cosmetic — CLAUDE.md requires no AI attribution in comments or PR bodies, and the MCP tool appended _Generated by [Claude Code]_ to all 15 comments. Those 15 are already corrected in place; this change stops it recurring. The instruction is bounded: read back, edit once, and if the edit re-appends the trailer, report it rather than looping.

What the run got right, unchanged here

Preferred --auto before a direct merge; re-ran discovery between merges; validated #90 in a throwaway worktree and cleaned it up; stopped at the 3-merge cap. It also correctly separated #130's Docker/Aspire test failure (infrastructure) from #124's GitHub.Copilot.SDK failure (a real build break), which is the distinction section 8 exists to make.

Validation

npm run check:docs passes.

The scheduled cloud runner is not a developer machine, and the prompt assumed
it was. Five corrections, all observed in the first MODE=merge run:

- `gh` is not installed there. The document is written in `gh` commands because
  they state clearly what to fetch, so keep them, but say up front that they may
  need translating: MCP tools for mutations, plain curl for bulk reads. The run
  worked this out unaided; that was luck, and luck is not a specification.
- curl reaches only repository-scoped endpoints. `/advisories` returns "sessions
  are bound to their configured repositories", so the security check now goes
  through `npm audit --package-lock-only`, which needs neither network nor auth.
  Read it both ways: an advisory against the incoming version blocks a merge, one
  against the outgoing version promotes an upgrade that was only Needs refresh.
- npm there is 10.9.7, so the allow-scripts note describes something that cannot
  happen. Scoped to npm 11+.
- The GitHub MCP comment tool appends an AI attribution trailer, which CLAUDE.md
  forbids. Posting is therefore not the last step: read the comment back and edit
  the trailer off. Fifteen comments from the first run have been corrected.
- Auto-merge refuses a PR that is already green, saying to merge directly. That
  is the expected answer, not a failure, and the document now says so rather than
  leaving a run to guess.
@github-actions

github-actions Bot commented Sep 1, 2026

Copy link
Copy Markdown
Contributor

🟢 Impact Analysis — PR #133

Risk tier: 🟢 LOW

📊 Summary

Metric Count
Files changed 1
Files added 0
Files modified 1
Files deleted 0
Modules touched 1

🎯 Risk Factors

  • 1 files changed (≤5 → LOW)
  • 1 module(s) touched (≤1 → LOW)

📦 Modules Affected

ci-workflows (1 file)
  • .github/prompts/dependabot-review.md

This report is generated automatically for every PR. See #733 for details.

@github-actions

github-actions Bot commented Sep 1, 2026

Copy link
Copy Markdown
Contributor

🛫 PR Readiness Check

ℹ️ This comment updates on each push. Last checked: commit c886074

PR Scope: 🔧 Infrastructure

⚠️ 1 item(s) to address before review

Status Check Details
Single commit 1 commit — clean history
Not in draft Ready for review
Branch up to date Up to date with dev
Copilot review No Copilot review yet — it may still be processing
Changeset present No source files changed — changeset not required
Scope clean No .squad/ or docs/proposals/ files
No merge conflicts No merge conflicts
Copilot threads resolved No Copilot review threads
CI passing All checks passing

Files Changed (1 file, +54 −1)

File +/−
.github/prompts/dependabot-review.md +54 −1

Total: +54 −1


This check runs automatically on every push. Fix any ❌ items and push again.
See CONTRIBUTING.md and PR Requirements for details.

@DeDuva
DeDuva merged commit 8560ac4 into dev Sep 1, 2026
16 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant