prepare-walkandlearn-input

name	prepare-walkandlearn-input
description	Legacy adapter that prepares a WalkAndLearn LangGraph input file and UI state JSON from a conversation copied to the macOS clipboard. Use only when the user explicitly asks for the legacy LangGraph workflow, LangGraph Studio/UI, or a WalkAndLearn graph state block. Do not use for Codex-native WalkAndLearn summarization; use generate-walkandlearn-summary instead.

Prepare WalkAndLearn Input (Legacy LangGraph)

Purpose

Retain the old LangGraph input workflow for explicit rollback use. Turn a clipboard conversation export into a local ignored input file for the W&L Summary graph, then copy the ready state JSON back to the clipboard.

For native Codex generation, stop and use generate-walkandlearn-summary.

Default target repo:

/Users/flo/Work/Private/Dev/AI/Agents

Workflow

1. Confirm the clipboard is plausible before writing:

   • Run pbpaste | wc -c.
   • Treat content under 2000 bytes as suspiciously short.
   • Do not print clipboard content in chat.

2. Resolve metadata before running the script:

   • Date must be concrete YYYY-MM-DD.
   • Resolve relative dates explicitly using the current date.
   • Topic should be a short slug-like phrase, e.g. rag-precision-recall.
   • If date or topic is missing, ask the user before running the script.
   • If topic is unclear, privately inspect only a small sample and suggest 1-3 concise topic slugs for the user to choose from.

3. Run the helper script:

python3 /Users/flo/.agents/skills/prepare-walkandlearn-input/scripts/prepare_walkandlearn_input.py \
  --date YYYY-MM-DD \
  --topic "topic words"

4. Report the generated filename and state block from script output.

Behavior Rules

• Never echo the conversation text back to the user.
• Generated files belong under agent_files/walkandlearn_summary/.
• Generated files are intentionally ignored by git; never force-add or commit them.
• The script overwrites the clipboard with the LangGraph state JSON.
• If LangGraph is not running at http://127.0.0.1:2024/ok, the script starts:

uv run langgraph dev

• LangGraph normally opens the UI itself; do not separately open the browser.
• If browser or Chrome control is available, best-effort focus an existing LangGraph tab after the script succeeds. Do not block the workflow if focusing fails.
• Do not create a thread or start a graph run through the API.

Script Options

• --date YYYY-MM-DD: Required session date.
• --topic TEXT: Required topic used for the filename slug.
• --repo PATH: Optional override for the Agents repo.
• --min-bytes N: Optional clipboard size threshold; default 2000.
• --no-launch: Do not start LangGraph if it is not running.
• --no-browser-focus: Marker for callers that browser focusing should be skipped.

Expected State Output

{
  "messages": [],
  "input_filename": "input-YYYY-MM-DD-topic-slug.md"
}