TL;DR
Recall can already help users find and revisit AI coding sessions. This proposal explores whether Recall should also help users reflect on their project-level AI coding workflow as a timeline of human inputs and agent outputs, so users can discuss what happened, notice repeated patterns, and decide whether to adjust their workflow, agents, or skills.
Problem
AI coding work often does not stay inside one clean session.
A user may copy context from one agent to another, continue the same task in a different tool, retry an approach later, or gradually turn repeated friction into a new workflow. Looking at sessions one by one can miss that larger process.
For reflection, the useful unit may not be “session”. It may be the project timeline.
Proposal
Add a project-level reflection mode that reconstructs a readable timeline across relevant sessions.
The default report should be conversation-first:
- user inputs;
- agent responses;
- important decisions, corrections, and turns;
- session titles or summaries when useful;
- timestamps and project/source context.
Low-level implementation logs should stay out of the main narrative by default:
- raw tool calls;
- file read/write logs;
- command execution details;
- verbose tool results;
- internal agent events.
Those details may still be useful as optional supporting context, but the main report should read like a clean history of human intent, agent response, correction, and outcome.
What this should help with
The reflection should help the user discuss questions like:
- Did work continue across multiple sessions or tools?
- Did the user repeatedly correct the same kind of agent behavior?
- Did the agent’s output from one session become the prompt or direction for another?
- Are there workflow habits that seem worth keeping, changing, or turning into a reusable skill/workflow?
- Is there a new custom workflow or agent behavior that would better match how the user actually codes?
The output should not present these as final judgments. It should surface patterns for discussion with the user.
Relationship to workflow improvement
This is related to the kind of manual reflection that led to projects like BDD Superpowers: observe recurring friction in AI coding, then encode better behavior into specs, plans, reviews, or skills.
The missing piece is that this currently depends on the user noticing the pattern manually. Recall already has the local session history that could help surface these patterns systematically.
The goal is not specifically “improve Superpowers”. It is broader:
- maybe the user should adjust an existing skill;
- maybe they should create a new workflow;
- maybe they should change the workflow for dispatching subagents;
- maybe they should keep the current process and just be aware of a recurring pattern.
Recall should help start that discussion.
Possible shape
For example:
recall reflect --project /path/to/repo
recall reflect --repo owner/repo --time 30d
Possible report sections:
-
Timeline
- A clean chronological narrative across sessions.
-
Observed patterns
- Repeated corrections, handoffs, retries, planning/debugging loops, or workflow friction.
-
Discussion prompts
- Questions for the user, such as:
- “Should this become a reusable workflow?”
- “Was this handoff intentional or accidental?”
- “Would a different planning/review checkpoint help here?”
-
Optional proposals
- Suggested workflow, agent, or skill changes that the user may accept, edit, or ignore.
TL;DR
Recall can already help users find and revisit AI coding sessions. This proposal explores whether Recall should also help users reflect on their project-level AI coding workflow as a timeline of human inputs and agent outputs, so users can discuss what happened, notice repeated patterns, and decide whether to adjust their workflow, agents, or skills.
Problem
AI coding work often does not stay inside one clean session.
A user may copy context from one agent to another, continue the same task in a different tool, retry an approach later, or gradually turn repeated friction into a new workflow. Looking at sessions one by one can miss that larger process.
For reflection, the useful unit may not be “session”. It may be the project timeline.
Proposal
Add a project-level reflection mode that reconstructs a readable timeline across relevant sessions.
The default report should be conversation-first:
Low-level implementation logs should stay out of the main narrative by default:
Those details may still be useful as optional supporting context, but the main report should read like a clean history of human intent, agent response, correction, and outcome.
What this should help with
The reflection should help the user discuss questions like:
The output should not present these as final judgments. It should surface patterns for discussion with the user.
Relationship to workflow improvement
This is related to the kind of manual reflection that led to projects like BDD Superpowers: observe recurring friction in AI coding, then encode better behavior into specs, plans, reviews, or skills.
The missing piece is that this currently depends on the user noticing the pattern manually. Recall already has the local session history that could help surface these patterns systematically.
The goal is not specifically “improve Superpowers”. It is broader:
Recall should help start that discussion.
Possible shape
For example:
Possible report sections:
Timeline
Observed patterns
Discussion prompts
Optional proposals