deepiri-sorge

Deepiri Sorge

Repo: deepiri-sorge — distributed AI PR review bot (GitHub Actions + optional Gemini/GitHub Models).

When to use this skill

• Opening or updating a PR that needs a first-pass AI review
• Wiring Sorge into a Deepiri repo (workflow + config + secrets)
• Interpreting Sorge comments or skip messages
• User says /sorge, "sorge", or "AI PR review"

Trigger a review (/sorge)

1. Ensure the PR exists and is pushed to GitHub.
2. Comment /sorge on the PR (issue comment body is exactly /sorge, optionally with a short note after).
3. Treat the bot reply as a first pass that informs — not replaces — manual review.
4. Fold Blocking / Important findings into the human review; do not merge on Sorge alone.

Also auto-runs on pull_request opened/synchronize/reopened (and configurable push) when the reusable workflow is installed.

Agent checklist (ship a PR)

• Push branch; open PR against main or dev as repo policy requires
• Comment /sorge on the PR
• Wait for the review comment (or skip notice)
• Address real issues; re-comment /sorge after large follow-up pushes if needed
• Complete human/CI review gates before merge

Install into a repo

1. Add .github/workflows/pr_review.yml calling the Sorge workflow (see deepiri-sorge README), or copy the workflow from that repo.
2. Optional root sorge.toml — defaults are fine for most repos.
3. Repo secret GOOGLE_API_KEY (Gemini) when using Gemini routing; GITHUB_TOKEN is provided by Actions.
4. Align with branch protection — Sorge comments are advisory unless the team explicitly requires a Sorge check.

Minimal sorge.toml

[sorge]
enabled = true

[filters]
min_lines = 20
skip_docs = true
skip_deps = true

[review]
style = "concise"
include_security = true
include_performance = true

Local dry-run (no PR comment)

cd deepiri-sorge   # or sibling clone from .axiom/ecosystem.json
poetry install     # or pip install -r requirements.txt
git -C <target-repo> diff origin/main...HEAD > /tmp/pr.diff
python -m bot.main --diff /tmp/pr.diff --dry-run --verbose