Evaluating a Pull Request

Evaluating a Pull Request

Run the ethical analysis scoped to a single PR: what deserves attention and what is worth discussing about what that PR introduces.

Quick version

conscience examine --pr https://github.com/your-org/your-repo/pull/42

URL formats

Both full URLs and short forms work:

# Full GitHub URL
conscience examine --pr https://github.com/your-org/your-repo/pull/42

# Short form
conscience examine --pr your-org/your-repo#42

The interval is the work, not the ticket

A PR analysis covers the time from the PR's earliest commit (or its opening, if that came first) to it being merged, closed, or now if it's still open. A PR opened and merged in six minutes still covers the hours of work that produced it.

Only AI activity inside that window is counted. A long-running session that straddles the window contributes the tokens, turns, and commands it produced during the PR, not its lifetime totals, so two PRs worked on in the same session get different numbers. The snapshot header and the PR comment both name the exact window.

What you get

The output starts with the PR's metadata:

  Conscience — PR #42: "Add user authentication"
  by alice | +681/−196 | 21 files | 3 review comments

Then the snapshot header and the standard analysis scoped to that PR's data. Add --full for every signal, the scorecard, and all reflection questions. Signals that commonly fire for individual PRs include:

• Prompt injection risk — if the PR body contains patterns that could manipulate AI tools
• Review engagement — whether the PR has review comments
• Contribution concentration — whether the PR's commits are from a single author

Adding AI session context

The current directory is the project by default. If your checkout lives elsewhere, pass --project:

conscience examine --pr your-org/your-repo#42 --project ~/code/your-repo

This correlates the PR with your AI sessions during its lifetime: token consumption, tools used, session duration, and signals like AI dependency ratio and tokenmaxxing.

JSON output

conscience examine --pr your-org/your-repo#42 --json

Emits the full snapshot. The signals, scorecard, and reflections keys are at the top level.

Posting the result on the PR

conscience examine --pr your-org/your-repo#42 --comment

Posts the analysis as a comment on the pull request, where reviewers already look. Two things to know:

• It's built from the sanitized export. A PR comment is visible to everyone with repository access, so it never carries file paths, shell commands, session ids, or names from signal evidence. Where evidence was a list of those, the comment shows a count. Signals about your conscience.yaml are left out too; they aren't about the change under review.
• A re-run edits the same comment. The comment carries a hidden marker, so ten pushes produce one conscience comment, not ten.

To see the body without posting:

conscience examine --pr your-org/your-repo#42 --markdown

Using in CI

The GitHub Actions workflow runs examine --pr ... --comment on every pull request. A CI runner has no AI session logs, so its comment covers the GitHub side and ends with the exact local command to add AI data; running that on your machine updates the CI comment in place. See Setting Up CI-CD for the full workflow.

Related

• Command Reference — full examine flags
• Security Signals — what prompt injection detection looks for
• The Seven Principles — what the scorecard measures