agentleFS
artificialbeingzartificialbeingz/abz-karpathy-llm-wiki/.claude/skills/save-session/SKILL.md

save-session

Skill1 starsChanged 6 months agoMITView on GitHub

Save a high-quality summary of the current session to raw/session-summary/<YYYY-MM-DD>-<slug>.md. Use when the user asks to "save session", "save the chat", "save our conversation", "session notes", or "summarize this session". The model writes the summary itself — there is no script to call. Output is a structured markdown file useful as a memory trail for future sessions, and as a candidate for ingestion into the wiki.

18 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use status and ship.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AnthropicSkill

    status

    Case status summary by audience — client-facing (plain language), internal (for the professor), or court-ready (formal caption format per local rules). Same facts, different framing and depth. Use when a student needs to update the client, brief the professor, or prepare a court status report.

    Also in 3 projects that use this

  • AWSSkill

    ship

    Finalize dev work on an issue branch — update docs, push, and open a PR.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 17 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Write a high-quality summary of this conversation to raw/session-summary/<YYYY-MM-DD>-<slug>.md (relative to the repo root).

    line 19
  2. Very specific:

    Receipts: when you mention a file, link it (path/to/file:line if relevant).

    line 73
  3. Distinctive:

    No filler: don't pad sections to look thorough.

    line 74
  4. Distinctive:

    No hallucinations: every claim must be traceable to the conversation.

    line 75

“Common” is measured: kept by at least 6 in 10 of the 17 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Skill: Save the current session

Write a high-quality summary of this conversation to raw/session-summary/<YYYY-MM-DD>-<slug>.md (relative to the repo root).

Steps

  1. Determine today's date — use date +%Y-%m-%d via Bash. Do not guess.
  2. Pick a slug — 2–4 lowercase hyphenated words capturing the session's focus. Examples: wiki-restructure, template-cleanup, decision-archive-pass. Ask the user if you can't infer a clear topic.
  3. Build the path — raw/session-summary/<date>-<slug>.md from the current repo root. Create raw/session-summary/ if it doesn't exist.
  4. Check if the file already exists — if a file with that exact name exists, ask the user whether to overwrite, append a section, or pick a different slug. Do not silently clobber.
  5. Write the summary using the template below. Pull facts only from the actual conversation — do not invent results, file paths, or commands that didn't happen. If a section has nothing to capture, omit it rather than padding.
  6. Confirm to the user with the file path and a one-line description of what was saved.
  7. Suggest ingestion if appropriate — if the session produced durable knowledge worth filing into the wiki (a new entity understood, a disambiguation resolved, a decision made), end with a one-line suggestion: "This summary contains [X]. Worth ingesting into the wiki?" If the session was routine, skip this step.

Template

# Session Notes — <YYYY-MM-DD> — <slug>

## Summary

<2–4 sentences. What was the user trying to accomplish, and what did the session deliver? Lead with the outcome, not the chronology. Mention the motivating question or trigger if there was one.>

## What was created

<Table of new files, one row each: path | purpose. Omit this section if no files were created.>

| File | Purpose |
|---|---|
| `path/to/file` | One-line description |

## What was changed

<Bullet list of edits to existing files. Group by file. Note WHY, not just WHAT, when the why isn't obvious from the file path. Omit if nothing was changed.>

- `path/to/file` — what changed and why

## Key facts captured (worth remembering)

<Bullets of non-obvious findings, decisions, or domain facts that would help a future session. Examples: surprising structural properties, naming conventions, deprecation status, integration boundaries, exact commands that worked. Skip conventions and architecture that can be re-derived from the repo. This section is the highest-value part — make it count.>

## Open questions / decisions deferred

<Anything left unresolved, ambiguous, or explicitly punted. Include the user's stated intent if it wasn't fully nailed down. Omit if everything was resolved.>

## Next steps planned

<Concrete follow-ups the user agreed to or that were queued. Skip speculative "could also do X" items.>

Quality bar

A good session note is one a future agent (or the user three weeks from now) can read cold and pick up the work. Aim for:

  • Specific over generic: "the deepest dependency chain is 4 hops: A → B → C → D" beats "the structure has deep nesting"
  • Why over what: "switched approach because the first one was failing on edge case X" beats "changed approach"
  • Receipts: when you mention a file, link it (path/to/file:line if relevant). When you mention a command, include the actual command.
  • No filler: don't pad sections to look thorough. Empty section = omit it. Three sharp bullets beat ten vague ones.
  • No hallucinations: every claim must be traceable to the conversation. If unsure whether something happened, leave it out.

Anti-patterns to avoid

  • Restating CLAUDE.md content (already in repo)
  • Listing every tool call ("then I ran ls, then I ran jq, then...")
  • Generic motivational summaries ("The session was productive")
  • Including conversation transcripts verbatim — this is a summary, not a log
  • Writing more than ~80 lines unless the session genuinely warrants it

Relationship to the wiki

This skill writes to raw/session-summary/ — the raw layer. Session summaries are not automatically wiki content. Per CLAUDE.md, most summaries are routine record-keeping; some teach durable lessons that should be ingested into the wiki. The decision to ingest is the user's, prompted by step 7 above.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.