Runs code reviews with the ocr CLI against an internal LLM gateway.
npx skills add https://github.com/peachest/skills --skill ocr-peachest
Please help me install this Agent Skill. Skill: ocr Source: https://github.com/peachest/skills/tree/main/in-progress/ocr Command: npx skills add https://github.com/peachest/skills --skill ocr-peachest
ocr (open-code-review) CLI so reviews succeed on the first attempt. ## Core Features & Use Cases - Gateway-safe review invocation: Enforces locked defaults (OCR_LLM_TIMEOUT, --timeout 30, effort-scaled concurrency) and routes scenarios like workspace diffs, single commits, branch/MR ranges, and whole-repo audits (ocr scan). - Provider configuration and verification: Configures the internal-gw provider correctly (extra_body under custom_providers, not llm.*) and ships a script that replays the real config against a fake server to prove thinking-disable fields reach the wire. - Session management and fix loop: Persists every review as a queryable session, supports resume and cross-session comparison, and hands critical/high/medium findings to /skill:fix with a low-effort re-review hook. - Use Case: A developer asks to review an MR touching 12 files. The Skill previews the diff, picks concurrency 4 with --effort low, runs the review in ~16 minutes, reports findings in Chinese grouped by severity, and offers to re-review after fixes land. ## Quick Start Review my current working copy changes with the ocr skill and report the findings grouped by severity.High-intent search queries and answers about installing and using this skill.
Run `OCR_LLM_TIMEOUT=300 ocr review --audience agent --timeout 30 --concurrency <n> --effort medium -b "<background>"`. Use `-c <sha>` for a single commit or `--from <base> --to <head>` for a branch or MR range.
Use `ocr scan --path <files-or-dirs>` with the same timeout and concurrency flags. Run `--preview` first to count files and set concurrency, and choose `--batch by-directory` for repo-layout-driven audits.
When `provider <name>` is active, ocr's resolver takes the provider path and ignores every `llm.*` setting, printing only a warning. The extra_body must be set at `custom_providers.<name>.extra_body` instead.
Common causes are thinking mode running on the wire (on GLM-5.3 only `reasoning_effort` works; `enable_thinking:false` is silently ignored), concurrency too high for the shared gateway, or a gateway-side hang producing 504s after a full timeout period.
Sessions persist in ~/.opencodereview/sessions/. List them with `ocr session list`, inspect with `ocr session show` or `ocr session comments`, and continue a range review with `ocr review --resume <session-id>`; there is no `ocr continue` command.
Judge scope with `--preview` first: 5 files or fewer allows concurrency 4, 6-15 files use 2-4, more than 15 files use 1-2, and whole-repo scans use 1. High concurrency on the shared gateway queues requests until they time out.
quiznatConfigure Hermes Agent CLI setup, providers, models, and tools.
malhashemiReview ticket documents for gaps, inconsistencies, and severity-classified issues.
fprochazkaAnalyzes GitLab merge request discussion threads and proposes fixes for unresolved comments.
mattothomasAnalyze PR diffs for SQL safety, LLM trust boundary violations, and side-effect risks.