Suggested test: copilot/pull-request-activation-base-sha-pinning #18367

Motivation

Link to the gh-aw PR: github/gh-aw#63499 — Pins the activation-job sparse checkout ref: to the PR base SHA (not PR head) when a workflow includes a pull_request trigger, preventing a same-repo PR author from influencing runtime-imported instructions/skills used to review their own PR (fixes github/gh-aw#63469).

Proposed test

• Workflow file: test-copilot-pull-request-activation-base-sha.md
• Trigger: pull_request (types: opened, synchronize)
• Engine: copilot
• Safe output: add-comment (simplest observable signal)
• Variant: standard

Minimal test prompt sketch

Author a workflow with a pull_request trigger that imports a runtime file (e.g. via imports: or an included instructions snippet) containing a sentinel string. In the test PR's head branch, modify that imported file to contain a different, "poisoned" sentinel. Have the agent report which sentinel value it observed via add-comment. Assert the comment shows the base-branch sentinel, proving the activation checkout used the PR base SHA rather than the head.

New fixtures or secrets needed

None — reuses existing create_test_pr/create_test_pr_with_branch harness helpers already used by test-copilot-close-pull-request.md and similar PR-triggered tests.

Notes

No existing pull_request-triggered test exercises runtime-import/activation-checkout content; existing PR-triggered tests (close-pull-request, update-pull-request, add-reviewer, mark-pull-request-as-ready-for-review) only check safe-output behavior on PR metadata, not activation-checkout provenance. Not present in fails.txt or open suggestions. Open question: confirming the simplest way to make the imported-file sentinel diverge between base and head without an actual security bypass attempt — may need a fixture file already present in main that the test PR modifies.

Generated by 🔍 Suggest New E2E Tests · copilot · auto · 65.1 AIC · ⌖ 9.43 AIC · ⊞ 8.5K · ◷