agentleFS
egregore-labsegregore-labs/egregore/.claude/skills/records/SKILL.md

records

Skill293 starsChanged 11 days agoMITView on GitHub

View or change your session records, hold the current session, redact a past record, record past sessions from this machine, and close or file the open items records carry.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    Use for /records, session record settings, the greeting's past-session offer, "off the record", a request to hold or redact a session …

    line 24
  2. Very specific:

    Use bin/records.sh; do not edit personal state directly.

    line 31
  3. Distinctive:

    Do not imply that hold removes earlier published revisions.

    line 66
  4. Very specific:

    When the greeting offer is answered: yes runs backfill --yes.

    line 68
  5. Very specific:

    If hold cannot identify the invoking session, report its one-line result.

    line 84
  6. Distinctive:

    Never guess a session id from a shared file or apply another person's id.

    line 85
  7. Distinctive:

    After fourteen days without an explicit reactivation, an unresolved item is

    line 104
  8. Distinctive:

    Use the outcome economics workflow when implementing findings from the scorecard.

    line 138

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

/records — your session records

Session records put asked, decided, left open, touched and cost in shared memory. Organization policy sets the default; each person controls their own records. The full trail stays on this machine as a file under ~/.egregore/trails/<folder>/<session>.md. Your own Claude account reads it once to write an outline. The model call can be disabled. See Session records for the exact content checks and their limits.

Claude Code, Codex CLI, Pi and Prime share the same record and hold controls. Pi and Prime record at session end; an idle sweep also covers Codex and continued sessions. /records sweep requests that sweep after the first offer has been answered. Codex reports tokens and price coverage; Pi and Prime use their usage ledger. Working-directory checks exclude other projects.

When to invoke

Use for /records, session record settings, the greeting's past-session offer, "off the record", a request to hold or redact a session record, or the open items the records carry: listing them, closing one, or filing one as an issue.

Commands

Use bin/records.sh; do not edit personal state directly.

IntentCommand
Show settings and unrecorded countbash bin/records.sh status
Read recent decisions and open itemsbash bin/records.sh digest
Read the organization's vitals, the last seven days beside the seven beforebash bin/records.sh vitals (--json for measurements and findings; --findings for actions and verification)
Record a concrete outcome and its contributing sessionsbash bin/records.sh outcome record --file <manifest.json>
Inspect accepted-outcome costs and coveragebash bin/records.sh outcome report --json
Register or update a measured correctionbash bin/records.sh intervention record --file <manifest.json>
Inspect the five/ten-outcome checkpointsbash bin/records.sh intervention report --json
Retry a waiting records pushbash bin/records.sh status --push
Turn your records on or offbash bin/records.sh on or off
Choose the detailbash bin/records.sh detail outline or detail full
Hold this sessionbash bin/records.sh hold
Hold a past session of yoursbash bin/records.sh redact <id-or-prefix>
Review past sessions before recordingbash bin/records.sh backfill --dry-run
Record the listed past sessionsbash bin/records.sh backfill --yes
Sweep idle sessions and refresh continued onesbash bin/records.sh sweep
Limit the past-session windowbash bin/records.sh backfill --days 30 --dry-run
Read one local trailbash bin/records.sh show <id-or-prefix>
Mark the greeting offer answeredbash bin/records.sh noticed
List active open itemsbash bin/records.sh open
Review retained stale itemsbash bin/records.sh open --stale
Renew the relevance of a verified existing outcomebash bin/records.sh reactivate <item-id>
Remove an item from current relevancebash bin/records.sh stale <item-id>
Mark an open item donebash bin/records.sh close <item-id>
Prepare an issue draft for an open itembash bin/records.sh file <item-id>
File that draft after the person confirms itbash bin/records.sh file <item-id> --confirm <request-id>

For "off the record", run hold immediately and confirm the returned result in one line. The durable hold applies across checkpoints, resumes and backfill. An existing record becomes a stub by default; a person who has turned off hold stubs gets a removal. Revisions already pushed remain in memory repository history. Purging history is a separate future command. Do not imply that hold removes earlier published revisions.

When the greeting offer is answered: yes runs backfill --yes; no keeps past sessions unpublished, and only sessions from now on are recorded; an explicit request to turn recording off runs off. Then run noticed. Only a later backfill --yes publishes the earlier sessions. For "show one", first run backfill --dry-run, choose one listed session, and use show to display its local trail. Return to the offer after the preview. Do not record the batch until the person says yes. A direct request to backfill is already that instruction.

full includes prompt excerpts and selected assistant text. Explain that before changing from outline when the person's intent is unclear. Outline asks are paraphrased with mechanical checks, while deterministic sections retain paths, command names, commit subjects and counts. The checks do not guarantee removal of every sensitive detail or outside name. Hold or redact is the person's remedy when something should stay out of shared memory.

If hold cannot identify the invoking session, report its one-line result. Never guess a session id from a shared file or apply another person's id. A past-session redaction is restricted to this person's records.

Digest

/records digest shows what the day's sessions decided and left open, including yesterday and today across all authors, and under Carried the open items still carried from earlier days. Each open item starts with its id. Use --days N or --since YYYY-MM-DD for a longer window. The digest reads records only and writes nothing. To keep a decision or finding in organizational knowledge, hand the line to /reflect.

Open items

An open item has two independent properties: resolution (open until an explicit close or filing) and relevance (active or stale). Its id remains the first six hex digits of the normalized line's SHA-256, without attribution. Its original age never resets. Similar wording is not proof of the same task.

After fourteen days without an explicit reactivation, an unresolved item is open but stale. This is a read-time projection; no timer rewrites records or appends expiry noise. Repeating the line does not renew relevance, and a mention cannot reopen a closed or filed item. Existing expired ledger events remain terminal for backward compatibility; no new expiry events are created.

/records open shows active work; open --stale shows the retained stale backlog; open --all --json exposes both with their original and relevance ages. The greeting shows the active count and a bounded stale count/link, without expanding stale text into the working context or demanding an owner.

/records stale <id> removes an item from current relevance immediately. /records reactivate <id> appends an explicit relevance renewal, retaining its original age and record. Reactivate only after checking that the same concrete outcome is still needed. If the goal or scope changed, close the old item with a successor reference in the next record and start a new concrete item. Never attach new work to a stale item merely because its broad topic sounds similar. Filed items continue in their linked issue, preventing duplicate carry work.

close, stale, reactivate and successful file append events to memory/sessions/open-items.jsonl. Events contain the item id, state, Runtime actor id, timestamp and optional issue URL, never item text. The latest valid event by time wins; the union merge rule preserves concurrent histories.

file without --confirm prepares a draft through the issue flow and files nothing. Show the person the draft it prints. Only after they confirm that exact draft, run file <item-id> --confirm <request-id> with the printed request id. The event is written only when the filing succeeds, with the issue link. Filing needs the configured GitHub issue tracker (issues.backend: github in egregore.json); without one, file says so and writes nothing.

Outcomes and measured corrections

Use the outcome economics workflow when implementing findings from the scorecard. The agent prepares manifests and records the work; no additional roles or approval queues are needed. Register a concrete outcome as open, link native aggregate session scopes, and append its verified disposition when the agreed acceptance condition is met. Do not infer acceptance from a merge, topic similarity or an open item. Do not invent a human verification. Keep attribution incomplete until driver, worker, retry and review costs are accounted for.

The explicit scorecard joins canonical final costs. Missing prices or records, held work and overlapping parent/worker scopes cannot become measured zero or complete unit cost. Repeated finding tags rank associated spend, not recoverable waste. Register an intervention's baseline before post-change work; inspect five comparable terminal outcomes, decide at ten, and revisit quality regressions immediately. These checkpoints never require delaying an already justified fix.

Telemetry

bash bin/telemetry.sh emit "command" '{"command":"records"}' 2>/dev/null &

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.