agentleFS
jelbirtjelbirt/LLM-Skill-Build/session-report/SKILL.md

session-report

Skill1 starsChanged 3 months agoApache-2.0View on GitHub

Write the status report this session sends back to the main session that delegated the work - the scope it was given, what it accomplished, exactly what changed on disk, how that was verified, what remains outstanding, and how the remainder connects to the tasks the receiving session is managing - in one fenced code block. Use whenever a delegated, side, or parallel session has to report up, or the user says "session report", "report back", "write up what you did", "status synopsis", "summarize this session for the main session", or "what do I paste back". Written for a reader that already has the project in context and merges into a session that keeps running, so use handoff instead when the next session starts empty.

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 4 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write the report THIS session sends back up to the session that delegated the work.

    line 9
  2. Distinctive:

    Every line should be something the reader could not already know.

    line 20
  3. Distinctive:

    Never let an unverified claim sit next to a verified one unmarked — a reader merging this into live context will act on both equally.

    line 59
  4. Distinctive:

    If this session was never given a scope — the user just started working — say that in Scope rather than inventing a mission after the fact.

    line 87
  5. Very specific:

    If there is no delegating session at all, say so and stop.

    line 89

“Common” is measured: kept by at least 6 in 10 of the 8 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

session-report

Write the report THIS session sends back up to the session that delegated the work. The reader is a running Claude Code session that already has the project in context — it does not need orienting, only updating.

Anything the user wants emphasized: $ARGUMENTS

What this is not

The reader already knows the project. Do not restate its shape, re-derive its architecture, explain decisions it made itself, or open with a preamble. No orient-first block — that is handoff's job, for a successor starting empty. Every line should be something the reader could not already know.

Instructions

  1. Re-check reality before writing. Report the repo, not your memory of it.

    • Uncommitted work: git status --short and git diff --stat HEAD.
    • Committed work needs a base commit, so establish one: the merge-base with the default branch (git merge-base HEAD origin/main, substituting the repo's actual default branch) when this session worked on its own branch, otherwise the newest commit in git log --oneline -20 that predates the session. Then git log --oneline {base}..HEAD and git diff --stat {base}..HEAD. If no base can be established, say that in the report rather than guessing at what this session committed.
    • git branch --show-current, un-pushed commits (git status -sb), and open PRs.
    • Anything created outside git — scratch files, generated artifacts, installed tooling — with absolute paths.

    A file you changed and forgot is the most damaging thing a report of this kind can omit.

  2. Compose the report with these sections, in this order. Keep every one — an empty section is stated as empty, never dropped, because a missing section reads as "nothing to report" rather than "not checked."

    • Scope — what this session was asked to do, in one or two lines, plus anything in that ask it deliberately did not do and why. When $ARGUMENTS carries the delegating ask itself, that text is the source of truth here; otherwise reconstruct it and say it was reconstructed.
    • Done — what was accomplished, against that scope. If the scope had five items and three landed, say so here rather than letting the reader infer it from what is missing.
    • Changed on disk — an explicit list of paths, one per line, each with a few words on what changed. Paths, not prose; taken from git in step 1, not from recollection. Include what was committed vs. left dirty, the branch, and anything written outside the repo. Say "no files changed" outright when that is the case.
    • Verified — for each claim of working, the evidence: the command run and its actual result. Anything not verified is listed as not verified. Never let an unverified claim sit next to a verified one unmarked — a reader merging this into live context will act on both equally.
    • Outstanding — what remains, each item labelled with why it remains: blocked (on what), deferred (by whose call), out of scope, or tried and rejected (with why). That last label is not filler — an approach this session ruled out is the one thing that stops the main session spending its own context rediscovering it.
    • Back to the main session — the connective tissue, and the reason this is a report rather than a summary: which outstanding items belong on the reader's task list, what it should stop doing or no longer assume, what moved under it (shared files, branches, schemas, state it also touches), and any decision now waiting on it, marked ⛳.
  3. Output the whole report in a single fenced code block, copy-ready, so it can be pasted into the main session in one motion. Use a four-backtick outer fence if the report itself contains triple-backtick fences. Do not write it to a file unless asked.
  4. After the block, add at most one line, marked ⛳ — and only for something the owner must act on personally, not the receiving session.

Notes

  • Terse beats complete. This lands in a context window that is already full; every line spent on what the reader knows costs a line it needed.
  • Report failures and dead ends as plainly as successes. A session that tried an approach and abandoned it saves the main session from retrying it, but only if the report says so.
  • If this session was never given a scope — the user just started working — say that in Scope rather than inventing a mission after the fact.
  • If there is no delegating session at all, say so and stop. Offer handoff when the next session starts empty, or a plain chat summary when the user just wants to know where things stand. Do not address a reader that does not exist.
  • Back to the main session comes last on purpose, even though it is what the reader acts on first. Writing it after the other five forces it to be derived from them rather than from memory, which is what keeps it honest.
  • Where the session's own narrative and git disagree, git wins. Say what the repo shows and flag the discrepancy.

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.