agentleFS
mrzlimrzli/skills/skills/conversation/collect/SKILL.md

collect

Skill0 starsChanged 7 days agoMITView on GitHub

Take in the notes yielded by the sessions below this one, at any depth, apply them to the current work, queue the changes agreed in them to implement now, and write the handoffs they call for. Use in a session whose forks, opened by spawn, have run yield.

7 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

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

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If an agreed item or a needed handoff conflicts with a decision made in this session since the fork was spawned, ask about it in one line …

    line 19
  2. Distinctive:

    Start each queued item once it is ready.

    line 33
  3. Distinctive:

    Every yield repeats the fork's whole picture, so check the latest file for that source ID first.

    line 39
  4. Distinctive:

    No file: write it.

    line 41

“Common” is measured: kept by at least 6 in 10 of the 3 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

Bring the results of the sessions below this one into the conversation. Each fork wrote its note with the yield skill.

Run agent-joins.sh collect. It says whether this session is a main session or a fork and gives its tree's ID, the start of the main session's session ID. It prints the notes from the sessions below this one, at any depth, that are new or changed since this session last collected, and records them as read. A note stays until every session above its fork has collected its latest version.

A note from a fork of a fork, marked "via" the forks between, uses the IDs of the fork just above it. Translate them through the map at the start of each of those forks' briefs, starting with the fork just above the note's fork and ending with this session's direct fork. That is the reverse of the order the "via" list shows. The briefs are at <git common dir>/agent-joins/briefs/NAME.md. An ID missing from a fork's map names an item new in that fork. Read it as that fork's name plus the ID.

Give each collected item an ID in this session's scheme, as the user's rules for IDs say, and keep its source ID beside it as a tracking key, such as G3 (Claude-2/A1a). Without such rules, refer to each item by its source ID alone.

Keep the reply very short: one line per collected note, saying only what happened, such as Claude-2 (FF1): resolved, design in docs/design/stock-exclusions.md; queued G3 (Claude-2/A1a); handoff G4 (Claude-2/A1b) at /tmp/handoff-trading-app-1a2b3c4d-Claude-2-A1b.md, or No new notes. Do not summarize the decisions. The notes are now in the conversation. Apply their decisions to the current work from here on, and refer to them when they become relevant.

Record durable decisions from the notes as the user's rules for project knowledge say. Without such rules, put them in the project's tickets or documentation, not only in the conversation.

If an agreed item or a needed handoff conflicts with a decision made in this session since the fork was spawned, ask about it in one line instead of queuing or writing it.

Agreed items

A note's "Agreed to implement" items were approved by the user in the fork, and each has a source ID, such as Claude-1/A1a.

In the main session, implement the items marked now through queue-work without asking for approval again. Add the items marked backlog to the project's backlog with their source IDs, and start none of them. Every yield repeats the fork's whole picture, so check each item's source ID against the queue and the backlog:

  • In neither: add it to the queue or the backlog, as its mark says.
  • In the queue or the backlog with the same decision and mark: skip it.
  • In the backlog, or in the queue and not yet started, with a changed decision or mark: update it there, and move it between the queue and the backlog when its mark changed.
  • In the queue with a changed decision, already started or integrated: add it as a new item.
  • In the queue, already started or integrated, with the same decision but now marked backlog: ask about it in one line.

Start each queued item once it is ready.

Handoffs

A note's "Handoffs needed" items each name a target repository and the change a handoff should ask for, and each has a source ID. In the main session, write each handoff with the handoff skill from its item, and give its path. Name the file after this repository's directory, the tree's ID, and the source ID, with / replaced by -, such as handoff-trading-app-1a2b3c4d-Claude-2-A1b.md. Fork names repeat across repositories and after a prune, so the repository and the tree's ID keep the files apart across repositories and main sessions.

Every yield repeats the fork's whole picture, so check the latest file for that source ID first. The first file has no number and counts as 1, and the latest is the one with the highest number:

  • No file: write it.
  • A file that asks for the same change: skip it.
  • A file that asks for a different change: write a new file with the next free number added, such as handoff-trading-app-1a2b3c4d-Claude-2-A1b-2.md, because a session in the other repository may already have used the old one. Say in its line that it changed.

In a fork

Implement nothing and write no handoffs, as spawn says for forks. Carry the agreed items and needed handoffs into the next note this fork writes with the yield skill.

Inactive trees

In any session, the script may report a tree as inactive for 14 days. Ask in one line whether to delete it, and on a yes run the script's prune command with the name of the tree's main session. Do not prune without the user's answer.

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.