agentleFS
geekxxgeekxx/ai-assistant-starter-kit/.claude/skills/full-sweep/SKILL.md

full-sweep

Skill1 starsChanged 4 months agoMITView on GitHub

The full Forward Deploy catch-up sequence, promoted from a memory-only procedure into an invocable skill. Runs, in order, a transcript sweep (all 6 sources), process-meetings, slack-recap, an Accenture calendar-forward check (detection only), sweep-jira (all 5 firms), sweep-binder (all 5 firms), and an action-item refresh — then a memory refresh. Trigger on "run a full sweep," "full sweep," "full catch-up." Never asks for scope; always covers everything since each lane's own last run.

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:

    Check all 6 sources per memory transcript-sweep-must-check-all-sources: Zoom Cloud Recordings (host), Zoom Notes recordings, Zoom …

    line 44
  2. Distinctive:

    Do not build a draft-forward path here — it hits the same external block and the value is the detection, not drafting near-identical …

    line 99
  3. Distinctive:

    Every lane above already stands alone and stays that way:

    line 139
  4. Very specific:

    Never skip a step in the sequence or reorder it — the ordering (transcripts → process-meetings → recap → Jira → Binder → action items → …

    line 153
  5. Very specific:

    Never auto-write a Jira transition, a gsheet cell, or a tracker update — those stay propose-then-approve inside sweep-jira / sweep-binder, …

    line 156
  6. Distinctive:

    Never forward, draft-forward, send, or add an attendee in the Step 3.5 Accenture check — it is detection only.

    line 159
  7. Distinctive:

    Never invent a combined-window state file.

    line 162
  8. Very specific:

    Never silently skip the Apple Reminders sync — it fires automatically once the action-item diff is approved (Step 6 here / Step 5.4 in …

    line 164

≈ “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
<!-- audience: internal -->

Full Sweep — On-Demand Orchestrator

Promotes the "full sweep" procedure (previously memory-only, confirmed with Jeff 2026-07-24 — see memory full-sweep-means-full-catchup) into a real, invocable skill. This is a convenience chain, not a new engine: every step below is an existing, independently-invocable lane. Running /full-sweep just runs them in the locked order so nothing gets skipped.

Reads _shared/fd-firms.md for the current firm list — passed through to each lane it invokes.

When to run

  • Jeff says "run a full sweep" or "full sweep."
  • Per memory full-sweep-means-full-catchup: do NOT ask for scope. "Full sweep" always means the whole sequence below, covering everything since each lane's own last run. If Jeff explicitly narrows it ("just the transcripts," "just a recap"), honor that instead of running the whole chain.

Window resolution — no new state file

There is no dedicated full-sweep state file. Each lane resolves its own window using the pattern it already implements — scanning owners_inbox/ (and owners_inbox/archive/) for the newest dated artifact of its own type:

  • process-meetings → newest meeting-notes_<date>_*.md
  • slack-recap → newest Slack Recap <date>.md
  • sweep-jira → newest jira-sweep_<date>_<firm>.md, per firm
  • sweep-binder → no dated-artifact window of its own; it diffs current Binder/tracker state against whatever the other lanes surfaced in this run

full-sweep does not invent a combined window — it just runs the lanes in sequence and lets each resolve its own, exactly as it would if invoked standalone.

Sequence (locked order — do not reorder)

1. Transcript sweep

Check all 6 sources per memory transcript-sweep-must-check-all-sources: Zoom Cloud Recordings (host), Zoom Notes recordings, Zoom Shared-with-me, Zoom Transcripts list, Slack, and email. Dedupe the same meeting appearing in more than one place (match on title + date + duration). Stage anything new to team_inbox/meeting_transcripts/. This step's own mechanics live in process-meetings Step 2 (2a–2d) and the shared _shared/zoom-transcript-retrieval.md fallback chain for non-host recordings — this orchestrator just makes sure the sweep happens before Step 2 below consumes it, per memory recap-after-process-meetings (process-meetings runs before the recap so transcript-derived notes inform it).

2. Invoke process-meetings

Runs its full procedure: builds the meeting index, gathers transcripts (including anything staged in step 1), dispatches a notes specialist per meeting, classifies audience + humanizes external notes, files notes and transcripts to the canonical FD Shared Drive, and reconciles action items (propose-then-approve) — including the Apple Reminders sync (Step 5.4) added to that skill, which fires automatically on the same approval, no second gate.

3. Invoke slack-recap (RECAP mode)

Invoke slack-recap in RECAP mode specifically — the catch-up digest, not the SWEEP (open-reply) mode that skill also offers. Runs both lanes (Slack + Outlook) over its own resolved window and files the recap to owners_inbox/. Runs after process-meetings so transcript-derived notes inform the recap (the locked ordering from memory recap-after-process-meetings). A full-sweep never triggers SWEEP mode; an open-reply sweep is a separate on-demand run.

3.5. Accenture calendar-forward check (detection only — never sends)

Jeff is Accenture-embedded and mirrors his Intuit meetings onto his Accenture calendar (jeffrey.heinen@accenture.com). Per memory intuit-external-autoforward-blocked, a standing auto-forward rule is impossible — Intuit blocks external auto-forwarding at the tenant level — so this stays a manual habit, and the sweep's job is to make sure nothing slips (the 8/17 Aprio-Construction + Diamond-standup meetings that were caught by hand are exactly what this prevents).

This step only detects and surfaces. It never forwards, drafts, sends, or adds an attendee — sending on Jeff's behalf is always his click.

Procedure:

  1. Window = since the last full-sweep run (the newest dated sweep artifact), matching the other lanes' since-last-run pattern. Only meetings added to the calendar since then are in scope, so already-handled meetings aren't re-flagged.
  2. List upcoming Intuit calendar events in that window (outlook_calendar list, times in PT per memory jeff-timezone-pt-seattle).
  3. For each, best-effort check whether it's already covered on the Accenture side, two ways: (a) jeffrey.heinen@accenture.com is already a native attendee on the invite (several of Jeff's recurring FD series already are — those need nothing), or (b) a forward to that address is found in Sent Items (outlook_email search over Sent). State the same honest caveat the reconcile research used: Sent-Items matching is unreliable (mutated subjects), so classify as [Covered] / [No forward found] / [Unclear] and never assert "not forwarded."
  4. Surface a short list — "New meetings to forward to Accenture" — of only the [No forward found] items that aren't already attendee-covered, with meeting name + date/time (PT). If all new meetings are covered, say so and add nothing. Jeff forwards them himself (or adds the attendee if he owns the meeting).

Do not build a draft-forward path here — it hits the same external block and the value is the detection, not drafting near-identical forwards.

4. Invoke sweep-jira for all 5 firms

Dispatched per firm (per _shared/fd-firms.md), in parallel. Produces the read-only jira-sweep_<date>_<firm>.md summaries automatically (no approval needed — read-only). Any PROPOSED-CHANGE it surfaces (an issue whose status should change per evidence from steps 1–3) is carried forward to the Step 6 approval round, not auto-applied.

5. Invoke sweep-binder for all 5 firms

Dispatched per firm, in parallel. Pass along what steps 1–4 surfaced — meeting notes, the Slack/Outlook recap, and the Jira sweep findings — as the evidence sweep-binder diffs the Binder/tracker against (see that skill's Step 3, "Accept inline evidence from other sweep lanes"). Produces a PROPOSED-CHANGES block per firm; nothing is written to a gsheet (no such tool exists — see sweep-binder's own gap note).

6. Refresh the consolidated action-item list

Same propose-then-approve pattern as process-meetings Step 4–5, now informed by everything steps 1–5 surfaced (meeting action items, recap action items, and anything from Jira/Binder that implies a task for Jeff). Present candidates for Jeff to pick — do not rewrite the list unasked. Exclude pure meeting-attendance items per memory action-items-exclude-attendance (keep only work that produces a deliverable, including meeting-prep work; fold posture/guardrail items into the nearest substantive item rather than dropping them). On approval, this reuses process-meetings Step 5.4's Reminders sync — do not build a second Reminders-sync path here.

7. Memory refresh (Sol, not a new tool)

Once Jeff approves the action-item and tracker changes, Sol updates the relevant per-firm engagement memory files and any other dated facts steps 1–6 surfaced (a new agent, a new blocker, a status change) so the next sweep starts from current state instead of rediscovering the same facts. This is a normal Sol memory-write, using the existing memory conventions already in place for this team (a frontmatter block with name/description/metadata.type, plus an index line added to MEMORY.md) — not a new tool or mechanism. Scope the refresh to what actually changed this run; don't rewrite a memory file that steps 1–6 didn't touch.

Independent invocability

Every lane above already stands alone and stays that way:

  • /process-meetings and /slack-recap are the existing skills — in spirit, the "transcript sweep" and "Slack/email recap" lanes respectively.
  • /sweep-jira [firm] and /sweep-binder [firm] (new, this build) run standalone against a single firm or all 5.

full-sweep is a convenience chain that runs all of them in the locked order — it is not the only way to run any one lane, and running one lane on its own (e.g. "just sweep Aprio's Jira") never requires going through full-sweep first.

What this skill must NOT do

  • Never ask for scope. "Full sweep" means the whole sequence, every time, covering everything since each lane's own last run. Only an explicit narrowing from Jeff ("just X") changes this.
  • Never skip a step in the sequence or reorder it — the ordering (transcripts → process-meetings → recap → Jira → Binder → action items → memory) is locked, matching memory full-sweep-means-full-catchup and recap-after-process-meetings.
  • Never auto-write a Jira transition, a gsheet cell, or a tracker update — those stay propose-then-approve inside sweep-jira / sweep-binder, unchanged by being run through this orchestrator.
  • Never forward, draft-forward, send, or add an attendee in the Step 3.5 Accenture check — it is detection only; forwarding is always Jeff's own click, and a standing external auto-forward is blocked by Intuit policy anyway (memory intuit-external-autoforward-blocked).
  • Never invent a combined-window state file. Each lane resolves its own window as it already does standalone.
  • Never silently skip the Apple Reminders sync — it fires automatically once the action-item diff is approved (Step 6 here / Step 5.4 in process-meetings), same approval, no second gate.

Related: memories full-sweep-means-full-catchup, full-sweep-includes-reminders, transcript-sweep-must-check-all-sources, action-items-exclude-attendance, recap-after-process-meetings; skills process-meetings, slack-recap, sweep-jira, sweep-binder; _shared/fd-firms.md, _shared/fd-sot-sources.md.

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.