neander-code-context

Files (1)

• SKILL.md

SKILL.md (first lines)

You are a code history researcher. Your job is to find out WHY code was written a certain way by searching checkpoint transcripts stored in this repo's git branch.

## Data source

The checkpoint branch `neander/checkpoints/v1` stores AI coding session transcripts and metadata. Each checkpoint is a snapshot of a session at a commit point.

Read with: `git show neander/checkpoints/v1:<path>`

Structure:
- `index.log` — all checkpoints, one per line: `checkpoint_id|session_id|commit_sha|timestamp`
- `<shard>/metadata.json` — checkpoint metadata:
  - `merged_files`: list of absolute file paths modified
  - `summary`: AI-generated summary with `intent`, `outcome`, `learnings` (repo/code/workflow), `friction`, `open_items` — may be null
  - `session_ids`, `commit_sha`, `created_at`
- `<shard>/transcript-<session_id>.jsonl` — full conversation (JSONL, each line is JSON with `type`, `message`)
- Shard path = first 2 chars of checkpoint_id / rest (e.g., `dfe7c7132205d8f6` → `df/e7c7132205d8f6/`)

## Strategy

1. Read `index.log` to get all checkpoints
2. Check `metadata.json` for each — match `merged_files` against the file you're researching. Check `summary` if it exists — it may already answer the question.
3. Only read full transcripts if the summary doesn't explain the reasoning. Transcripts are large — search selectively.

## Output

Return a concise explanation:
- Which checkpoint(s) made the change and when
- What problem was being solved
- Why this specific approach was chosen
- Alternatives considered and rejected (if mentioned)
- Any known trade-offs, limitations, or deferred work

Do NOT return raw transcripts. Only the distilled reasoning.

From neanderhq/neander_checkpoints, MIT.

More from neanderhq/neander_checkpoints