Skip to content

[Feature] Persist critique history across sessions and providers #128

@DebZakari

Description

@DebZakari

Hi! I’m new to Impeccable, so it’s possible I missed an existing workflow. I looked through the docs and current behavior, and I couldn’t find a built-in way for /impeccable critique to persist previous review findings across sessions or coding agents.

What problem does this solve?

Currently, /impeccable critique appears to work as a one-off review. After I address the findings and run critique again in a new session or with a different agent, the previous critique context is not available unless I manually paste it back in.

This creates a few workflow issues:

  • The agent may re-raise items that were already addressed or intentionally deferred.
  • It may introduce a fresh set of lower-priority issues without understanding the previous review trajectory.
  • Scores are harder to interpret over time because there is no persistent critique baseline.
  • The current workaround is to manually copy/paste previous critique output into the new session, which is tedious in terminal-based workflows.

Proposed solution

Add an optional persisted critique file, for example CRITIQUE.md, that /impeccable critique can read and update.

Possible behavior:

  • If CRITIQUE.md exists, /impeccable critique reads it before generating a new review.
  • After each critique, the agent updates the file with:
    current score
    previous score
    unresolved issues
    resolved issues
    deferred or intentionally accepted tradeoffs
    recommended next commands
  • Users can opt in to this behavior, rather than making critique persistence mandatory.
  • Users can reset or archive CRITIQUE.md when they want a fresh review.

This would make critique feel more iterative, especially across sessions and providers.

Provider(s) this applies to

  • Cursor
  • Claude Code
  • Gemini CLI
  • Codex CLI
  • VS Code Copilot
  • Kiro
  • OpenCode
  • Qoder
  • All providers

Alternatives considered

I tried manually creating a CRITIQUE.md file and asking the agent to use it as context for follow-up critique passes. That worked surprisingly well, but it requires manual setup and is easy to forget.

A built-in convention would make the workflow much smoother and more consistent.

Metadata

Metadata

Assignees

Labels

enhancementNew feature or request

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions