Aggregates ORC behavior traces into calibration reports filed as upstream GitHub PRs.
npx skills add https://github.com/azure-id/orc --skill orc-retro-azure-id
Please help me install this Agent Skill. Skill: orc-retro Source: https://github.com/azure-id/orc/tree/main/templates/skills/orc-retro Command: npx skills add https://github.com/azure-id/orc --skill orc-retro-azure-id
.jsonl sidecars, and aggregates per-band retries, requeues, downgrades, and gate bounce rates. - Read-only, report-only operation: Never edits the rubric, skills, config, or project code; it dispatches mining to a cheap subagent and validates the return against a strict contract. - Upstream delivery: Files the AI-readable report (YAML frontmatter plus human sections) as a PR, with issue fallback, to the configured retro_repo via the gh CLI or a GitHub MCP; a P0 preflight stops the run entirely if no delivery channel exists. - Use Case: After a week of ORC runs, ask how well the scoring is calibrated. The Skill mines all traces, reports per-band outcomes and two silent model downgrades, and opens a PR titled orc-retro: 130726 — 3 runs, 2 recommendations against the ORC repo. ## Quick Start Run /orc-retro to analyze the logged ORC traces and file a calibration report to the ORC repository.High-intent search queries and answers about installing and using this skill.
Run /orc-retro after ORC runs have written traces to the configured log_dir. The Skill collects the .txt traces and .jsonl sidecars, dispatches a mining subagent, and returns per-band statistics, downgrades, and recommendations with evidence line references.
The report is filed upstream to the configured retro_repo (default azure-id/orc) as a pull request, with issue creation as fallback. Delivery uses the gh CLI when authenticated, or a GitHub MCP server; if neither channel exists, the retro stops before mining anything.
No. The Skill is strictly read-only and report-only against the local system. It never edits effort-and-mode.md, skills, config, or code; recommendations are phrased for a human or the ORC repo maintainer to apply.
The P0 preflight requires a delivery channel: an authenticated gh CLI or a GitHub MCP server. If neither exists, the run stops immediately and tells you to run gh auth login or connect a GitHub MCP, since an undeliverable report is pointless.
Every aggregate states its sample size n, and any recommendation based on fewer than three runs is labeled weak signal rather than stated as a conclusion. With zero traces, the Skill reports that no runs have been traced and stops without inventing findings.
Foreign dispatches appear as EXTRA lines with no SPAWN or RETURN, so the Skill runs orc extra stats --json instead of parsing them. It reports outcome mix, substitution and reroute counts per profile per band, and never treats a missing RETURN as a leak.
GuicedEEIdentify root causes of bugs using a five-step debugging workflow.
parhummAggregate test results, security findings, and PR feedback into a composite score with routing recommendation.
kok1eeeUpdate SKILL.md Gotchas from execution history and memory sources.
HomericIntelligenceParse and categorize CI failure logs to identify root causes.