DevAgentRadar → Repo Harness Rule Patch

Goal: update our repository's permanent harness instructions based on this release.

Update the file that tool actually reads. Do not dump everything into one file.

• Claude Code → CLAUDE.md plus .claude/ (skills, hooks, settings, agents, commands). It does not read AGENTS.md natively; start CLAUDE.md with @AGENTS.md.
• Codex, Copilot, Cursor, Factory Droid, Grok Build, Roo Code, Goose, OpenCode, Amp, Zed, Aider → AGENTS.md.
• Gemini CLI / Antigravity → GEMINI.md. AGENTS.md only if context.fileName is set.
• Cursor glob-scoped rules → .cursor/rules/*.mdc (plain .md is ignored), not a second constitution.
• Codex MCP → .codex/config.toml, not .mcp.json.
• Copilot extra instructions → .github/copilot-instructions.md; custom agents → .github/agents/.
• Windsurf → .windsurf/rules (do not assume AGENTS.md).
• Cline → .clinerules.
• Procedures → a skill (SKILL.md). Enforcement the model must not skip → a hook. Isolated roles → subagents.
• Do not fork the same rule into three tool files.

Context

Assistant: Kimi Code Proposal: Rewrite plugin skill invocations to qualified pluginId:skillName names Summary: Kimi Code 2.3.0 requires plugin skills as <pluginId>:<skillName> (bare names no longer work), expands [permission] dangerous_command_guard = false to disable all confirmation guards, moves session-rating opt-out from tui.toml to config.toml, and adds a model-switch dialog plus queued /compact. Primary source: https://github.com/MoonshotAI/kimi-code/releases/tag/%40moonshot-ai/kimi-code%402.3.0

Why it matters

Kimi Code 2.3.0 lists and invokes plugin skills only as <pluginId>:<skillName>, so same-named skills from different plugins stay visible; a bare skill name no longer works, and any prompt, slash command, or doc still using one will not invoke the plugin skill. The same release widens [permission] dangerous_command_guard = false so it turns off every built-in confirmation guard—dangerous commands, sensitive-file access, and git control-path access—not only the dangerous-command guard, which makes existing quiet configs broader than before. Session rating opt-out moved from tui.toml to config.toml. When cloud configuration enables it, switching models in a session with history offers start a new session, compact and continue, continue directly, or cancel; /compact now joins the message queue if the agent is busy and is skipped if auto-compact ran while it waited. With experimental tool-select on, [mcp] default_deferred = true loads every MCP server's tools on demand. Align skill names, permission flags, rating config, and compact/switch choices so 2.3.0 sessions keep working without dropping guards or context.

Suggested actions

1. Replace every Kimi Code plugin skill invocation in this repo (prompts, slash commands, READMEs, and agent configs) with the qualified name <pluginId>:<skillName>; Kimi Code 2.3.0 no longer runs plugin skills by bare name, so same-named skills from different plugins stay visible and invocable.
2. If config.toml has [permission] dangerous_command_guard = false, remove that line or set it to true unless you intentionally want Kimi Code 2.3.0 to turn off all built-in confirmation guards (dangerous commands, sensitive-file access, and git control-path access).
3. Move the session-rating opt-out from tui.toml into config.toml and delete the tui.toml copy; Kimi Code 2.3.0 reads that setting only from config.toml.
4. When the experimental tool-select flag is on, add [mcp] default_deferred = true to config.toml so every MCP server loads tools on demand; set deferred: false only on servers that must not be deferred.
5. On a model switch in a session with history, choose compact and continue (or send /compact, which now queues while the agent is busy and is skipped if auto-compact already ran) instead of continue directly; start a new session or cancel if the existing history should not follow the new model.

Config surfaces this release may change

• skills — check your repo before applying
• permission rules — check your repo before applying
• hooks — check your repo before applying
• MCP servers — check your repo before applying
• context and compaction settings — check your repo before applying
• subagent definitions — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/kimi-code-moonshot-ai-kimi-code-2-3-0-rewrite-plugin-skill-invocations-t and mark Applied, Skipped, or Failed. Proposal id: 6d831f41-08d0-473e-bc37-fdd7f6ff26fe

Instructions:

1. Read the existing context files and settings for the tools in this repo.
2. Draft an explicit Git diff. Update ONLY agent harness rules or tool configuration.
3. Call out deprecated flags, obsolete habits, or changed permission boundaries.
4. Do not touch application logic.

Start with the proposed diff and say which layer and which tool's file it belongs in.