The "never resolve proactively" rule lives in `integrations//skills/crit-cli/SKILL.md` (claude-code, cursor, codex, github-copilot, opencode), but it's:

1. Missing from the review-skill family — `integrations//skills/crit/SKILL.md` mentions replying with `crit comment --reply-to ...` but doesn't tell the agent NOT to add `--resolve`.
2. Missing from the single-file integrations — `integrations/cline/crit.md` and `integrations/aider/CONVENTIONS.md` have no resolve guidance at all.
3. Buried where it does exist — in `crit-cli/SKILL.md` it sits at line ~76 inside a wall of text, easy for an agent to miss while scanning.

Symptom

Real failure today: agent was given a one-shot prompt (`crit fetch` → reply → share workflow) that didn't mention `--resolve` either way. The `crit-cli` skill wasn't auto-invoked for the share workflow, so the agent fell back on training-data habit and added `--resolve` to both replies. Resolving is the reviewer's call, not the responder's.

Proposed fix

• Add the anti-resolve sentence to every `crit/SKILL.md` (review skill) + cline + aider, near the `--reply-to` example.
• Promote the rule in `crit-cli/SKILL.md` from line ~76 to the top of the "Replying to comments" section so it's the first thing the agent sees.
• Bump the version in `integrations/claude-code/.claude-plugin/plugin.json` (and any sibling manifests with version fields) once shipped, so users actually get the updated prompt.

Files to touch (~7):

• `integrations/claude-code/skills/crit/SKILL.md`
• `integrations/cursor/skills/crit/SKILL.md`
• `integrations/codex/skills/crit/SKILL.md`
• `integrations/github-copilot/skills/crit/SKILL.md`
• `integrations/opencode/SKILL.md`
• `integrations/cline/crit.md`
• `integrations/aider/CONVENTIONS.md`
• All `crit-cli/SKILL.md` files (reorder)