From autoreview
Use when user wants to adapt a remote rule's tier / triggers / severity / type WITHOUT forking the rule. Wizard for editing config.yaml's `remote_rules[].overrides` block. Triggers include "this corp rule is too noisy in my repo", "downgrade this rule to warning", "narrow this rule to only src/", "make this rule manual-only". Skip when the rule is local — local rules are owned by the user; edit the file directly. Skip when user wants to change rule body — that's a fork, not an overlay.
How this skill is triggered — by the user, by Claude, or both
Slash command
/autoreview:override-ruleThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> **Cross-platform.** Snippets below use bash-style env-var syntax (`${CLAUDE_PLUGIN_ROOT}`). Claude Code's Bash tool runs Git Bash on Windows so these work as-is; on native PowerShell substitute `$env:CLAUDE_PLUGIN_ROOT`, on cmd use `%CLAUDE_PLUGIN_ROOT%`. Plugin requires Node ≥22 — only assumed binary.
Cross-platform. Snippets below use bash-style env-var syntax (
${CLAUDE_PLUGIN_ROOT}). Claude Code's Bash tool runs Git Bash on Windows so these work as-is; on native PowerShell substitute$env:CLAUDE_PLUGIN_ROOT, on cmd use%CLAUDE_PLUGIN_ROOT%. Plugin requires Node ≥22 — only assumed binary.
Seven steps. Overlays let a repo tune remote (corp-shipped) rules without forking. Changes live in remote_rules[<name>].overrides[<rule-id>] in .autoreview/config.yaml (or config.personal.yaml for personal-scope changes).
Only frontmatter is overridable (name, triggers, tier, severity, type, description). If the user wants to change the rule body, that requires forking the rule into a local copy — explain this and stop.
Pick remote rule. List rules grouped by source. Show rule ids, names, and current effective tier/severity. Ask user to pick one by id.
Example listing:
corp-standards:
corp/audit-log-on-handlers tier: critical severity: error
corp/no-todo-without-ticket tier: standard severity: error [manual]
vendor-pack:
vendor/naming-conventions tier: trivial severity: warning
Show effective frontmatter. Read the upstream frontmatter from .autoreview/remote_rules/<name>/<ref>/<path>/<rule>.md and merge any existing override from config. Display both the raw upstream values and the currently applied overrides side by side:
Field Upstream Current override Effective
--- --- --- ---
tier critical trivial trivial ←override
severity error (none) error
type auto manual manual ←override
triggers dir:"src" (none) dir:"src"
Decide what to change. Ask which field(s) to adjust. Overridable: tier, triggers, severity, type, description, name. Not overridable: body (explain why and stop if that's what they want).
If triggers changed — breadth check. Run node ${CLAUDE_PLUGIN_ROOT}/scripts/bin/check-breadth.mjs --expr '<new-expr>' against the new trigger expression. Zero LLM cost. Iterate until the match set looks right. A trigger that matches 0 files is dead code; a trigger that matches too many files will inflate review cost.
If tier changed — test-drive. Read .autoreview/config.yaml and explain the cost/time tradeoff at the new tier:
| Tier | Typical use | Relative cost |
|---|---|---|
trivial | Style, naming, simple patterns | Cheapest / fastest |
default | General conventions | Baseline |
standard | Cross-cutting concerns, API contracts | Moderate |
heavy | Architectural rules, security invariants | Slow / expensive |
critical | Compliance, audit, zero-tolerance violations | Most thorough |
If the target tier is not defined in the repo's tiers: config, walk the user through adding it (same as step 5 of autoreview:create-rule). Then run node ${CLAUDE_PLUGIN_ROOT}/scripts/bin/reviewer-test.mjs --rule <id> --file <path> against 2–3 sample files at the new tier so the user can see how the new model behaves before committing.
Write overlay. Run the save command, choosing scope:
--scope repo → writes to .autoreview/config.yaml (committed, shared with team).--scope personal → writes to .autoreview/config.personal.yaml (gitignored, local only).Ask the user which scope to use if not obvious from context. Get explicit user consent before writing.
node ${CLAUDE_PLUGIN_ROOT}/scripts/bin/override-rule.mjs save \
--remote <source-name> \
--rule <rule-id> \
--field tier=trivial \
--field severity=warning \
[--scope personal]
Multiple --field flags are accepted. Field format is <name>=<value>. To explicitly unset an override field (revert to upstream value), pass <name>=null.
Confirm + advise. Print the saved overlay block. Suggest next steps:
autoreview:review to verify the overlay applied as expected.autoreview:context <some-path> to confirm which files now match.config.yaml if scope was repo.remote_rules:
- name: corp-standards
url: https://github.com/acme/autoreview-rules
ref: v1.2.0
path: rules
overrides:
audit-log-on-handlers:
tier: trivial # downgraded from corp's critical
severity: warning # non-blocking in this repo
no-todo-without-ticket:
type: manual # opt-in only; don't run on every commit
| Situation | Action |
|---|---|
| User wants to change the rule body | Explain that body is not overridable. Offer to fork: copy the upstream .md file into .autoreview/rules/ under a local id, then edit freely. |
Rule is local (id has no <source>/ prefix) | Local rules are owned by the user — edit the file directly. |
| User wants to override a field not in the allowed list | Tell them the allowed fields: name, triggers, tier, severity, type, description. |
| Tier not defined in config | Walk through adding it (step 5) before saving. Don't save an overlay referencing an undefined tier. |
npx claudepluginhub krzysztofdudek/autoreview --plugin autoreviewProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.