Description
Add a partio improve command that analyzes captured session transcripts from checkpoints to identify recurring friction patterns (repeated errors, tool-call retries, misunderstandings, wasted turns) and generates actionable suggestions for improving context files like CLAUDE.md.
The command should use a two-phase approach:
- Index phase: Scan checkpoint session data to identify recurring friction themes (e.g., agent repeatedly hitting the same linting error, misunderstanding project conventions, retrying failed approaches).
- Suggest phase: Read relevant transcript excerpts and generate specific improvement suggestions — with evidence quotes and proposed diffs — for context files that would prevent the friction in future sessions.
This leverages Partio's unique position of having captured session transcripts to close the feedback loop: sessions reveal what the agent struggled with, and those struggles inform better prompts/context for future sessions.
Why
Partio already captures the full reasoning behind code changes. This feature turns that captured data into direct value by helping users write better CLAUDE.md files and project instructions. Without this, users must manually review session logs to identify patterns — a tedious process that rarely happens. Automated friction analysis makes the captured session data actively useful rather than purely archival.
Source
Target Repos
Acceptance Criteria
Description
Add a
partio improvecommand that analyzes captured session transcripts from checkpoints to identify recurring friction patterns (repeated errors, tool-call retries, misunderstandings, wasted turns) and generates actionable suggestions for improving context files likeCLAUDE.md.The command should use a two-phase approach:
This leverages Partio's unique position of having captured session transcripts to close the feedback loop: sessions reveal what the agent struggled with, and those struggles inform better prompts/context for future sessions.
Why
Partio already captures the full reasoning behind code changes. This feature turns that captured data into direct value by helping users write better CLAUDE.md files and project instructions. Without this, users must manually review session logs to identify patterns — a tedious process that rarely happens. Automated friction analysis makes the captured session data actively useful rather than purely archival.
Source
entireio-cli-pullsTarget Repos
cliAcceptance Criteria
partio improveanalyzes recent session transcripts and identifies recurring friction patterns (repeated errors, retries, misunderstandings)partio improvegenerates specific suggestions for CLAUDE.md improvements with evidence excerpts from transcripts