refactor(skill): make proceed-with-the-recommendation agent-agnostic#38
refactor(skill): make proceed-with-the-recommendation agent-agnostic#38naimkatiman merged 1 commit intomainfrom
Conversation
Replace Claude-specific phrasing with neutral "agent" terminology across the skill, plugin mirror, command, and README so the skill is invocable from any AI coding agent (Claude, Codex, Gemini, etc.) and the precondition rule covers any agent-instruction file (CLAUDE.md / AGENTS.md / GEMINI.md / equivalent).
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request generalizes the documentation and skill definitions by replacing specific references to "Claude" with "the agent" and expanding the list of recognized instruction files to include AGENTS.md and GEMINI.md. The review feedback suggests using the singular AGENT.md naming convention and explicitly mentioning "rules" to ensure the agent correctly interprets the instruction to check for conflicts within those files.
| - Recommendations include destructive actions (deploy, force-push, DB drops, secret changes) without prior explicit authorization | ||
| - User scoped the work ("just the first one", "only the safe ones") — honor the scope | ||
| - Recommendations conflict with project CLAUDE.md rules | ||
| - Recommendations conflict with project agent-instruction files (CLAUDE.md / AGENTS.md / GEMINI.md / equivalent) |
There was a problem hiding this comment.
The phrasing "agent-instruction files (... / equivalent)" could be slightly ambiguous for an AI agent. Using "rules" (as in the original text) and clarifying "or equivalent" helps ensure the agent understands it should check the content of these files for conflicting instructions. Also, AGENT.md is a more common singular convention than AGENTS.md when referring to a specific agent's instruction file.
| - Recommendations conflict with project agent-instruction files (CLAUDE.md / AGENTS.md / GEMINI.md / equivalent) | |
| - Recommendations conflict with rules in project agent-instruction files (CLAUDE.md / AGENT.md / GEMINI.md / or equivalent) |
| - Recommendations include destructive actions (deploy, force-push, DB drops, secret changes) without prior explicit authorization | ||
| - User scoped the work ("just the first one", "only the safe ones") — honor the scope | ||
| - Recommendations conflict with project CLAUDE.md rules | ||
| - Recommendations conflict with project agent-instruction files (CLAUDE.md / AGENTS.md / GEMINI.md / equivalent) |
There was a problem hiding this comment.
The phrasing "agent-instruction files (... / equivalent)" could be slightly ambiguous for an AI agent. Using "rules" (as in the original text) and clarifying "or equivalent" helps ensure the agent understands it should check the content of these files for conflicting instructions. Also, AGENT.md is a more common singular convention than AGENTS.md when referring to a specific agent's instruction file.
| - Recommendations conflict with project agent-instruction files (CLAUDE.md / AGENTS.md / GEMINI.md / equivalent) | |
| - Recommendations conflict with rules in project agent-instruction files (CLAUDE.md / AGENT.md / GEMINI.md / or equivalent) |
Replace Claude-specific phrasing with neutral "agent" terminology across gateguard, token-budget-advisor, verification-loop, strategic-compact, and the source skills README so these skills are invocable from any AI coding agent (Claude, Codex, Gemini, etc.). Source files (5) + plugin mirror (5, regenerated via npm run build). Phrasing-only refactor; no behavior change. 238/238 tests pass. Out of scope: - proceed-with-the-recommendation (covered by PR #38) - workspace-surface-audit (most matches are legitimate product references; deferred to a separate audit) - ralph (both Claude references are legitimate: section header for Claude Code usage and the literal scripts/ralph/CLAUDE.md path) - token-budget-advisor.md attribution line ("Token Budget Advisor for Claude Code" is the original project name)
1 participant
Summary
proceed-with-the-recommendation(skill, plugin mirror, command, README) to neutral "agent" terminology so the skill is invocable from any AI coding agent (Claude Code, Codex, Gemini, etc.).Files
commands/proceed-with-the-recommendation.mdplugins/continuous-improvement/commands/proceed-with-the-recommendation.mdplugins/continuous-improvement/skills/README.mdplugins/continuous-improvement/skills/proceed-with-the-recommendation/SKILL.mdskills/proceed-with-the-recommendation.mdOut of scope (deliberately left alone)
Restart the Claude Code sessioninstall line stays — install path is genuinely Claude-Code-specific even if the skill content is portable. A future PR can add per-agent install sections if needed.Test plan
npm test: 238/238 pass