agentleFS
geekxxgeekxx/ai-assistant-starter-kit/.claude/skills/process-meetings/SKILL.md

process-meetings

Skill1 starsChanged 4 months agoMITView on GitHub

On-demand meeting-to-action-items workflow. Builds a meeting index from the Outlook calendar (all attended meetings), reads .vtt transcripts from the drop-folder (primary) and pulls Zoom cloud recordings for meetings Jeff hosts (secondary), flags indexed meetings with no transcript as gaps, dispatches a specialist per meeting to produce structured notes filed to Google Drive, humanizes notes for Op Mechs and firm meetings (see `_shared/fd-firms.md` for the current firm list) since Jeff posts those to Slack, then reconciles new action items into the consolidated list — proposing changes for Jeff's approval before editing it, then syncing approved items into Apple Reminders. Use when Jeff says "process my meetings," "turn my transcripts into notes," "catch up my action items from meetings," or similar.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. davekilleen/Dex★ 509same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository13 other files it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. About 0 are common, 7 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Default window = since the last processed meeting:

    line 50
  2. Distinctive:

    Fallback: if none exists, use the previous 48 hours.

    line 54
  3. Distinctive:

    If Jeff named a window or a specific meeting, that overrides.

    line 55
  4. Very specific:

    Call outlook_calendar_search with query: "*", afterDateTime/beforeDateTime set to the window bounds.

    line 75
  5. Distinctive:

    Only meetings Jeff hosts appear here.

    line 96
  6. Distinctive:

    Follow that file's SSO pause-and-ask-Jeff rule and its fail-loudly failure modes — a meeting only becomes a Step 2d gap after the chain …

    line 130
  7. Very specific:

    If a meeting appears in both the drop-folder and Zoom, prefer the .vtt (it's the source Jeff curated).

    line 135
  8. Distinctive:

    Do this after the notes come back and before routing them to Drive (Step 3.6), so the version filed to record is the finished one.

    line 202
  9. Very specific:

    Run humanizer over the note body — remove em/en dashes, emojis, rule-of-three, inline-header bullet lists, promotional filler, and the …

    line 217
  10. Very specific:

    Do not humanize internal notes.

    line 230
  11. Very specific:

    Do NOT silently write to a stale path.

    line 249
  12. Very specific:

    Read the current consolidated list — the newest owners_inbox/consolidated-action-items_<date>.md.

    line 265
  13. Very specific:

    If a [FD #N] numbering collision or ambiguity comes up (e.g.

    line 318
  14. Very specific:

    Use safe-mv.sh (--dry-run first, then real).

    line 341
  15. Distinctive:

    One meeting = one note = one subagent: Don't merge two meetings into one note.

    line 373
  16. Very specific:

    Non-host transcripts: use the shared fallback chain in _shared/zoom-transcript-retrieval.md (Zoom API → share-link token via the browser …

    line 379

≈ “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

Process Meetings — On-Demand

Turns Zoom meetings into structured notes and keeps the consolidated action-item list current, as one repeatable procedure. This replaces the ad hoc manual flow (export .vtt → hand-write notes → separately reconcile items) that let filed notes slip past the action-item list. Sol orchestrates and reviews; a specialist does the synthesis. Sol never writes the notes directly.

Design decisions (locked with Jeff 2026-07-10):

  • Drop-folder is primary. Jeff exports .vtt transcripts to team_inbox/meeting_transcripts/ — this is the main path, because most of his meetings are shared with him, not hosted by him (see below). The Zoom API fetch is a secondary path that only covers meetings Jeff hosts (the Zoom recordings_list/search_meetings endpoints only surface a user's own cloud recordings, never shared-with-me ones — verified live 2026-07-10).
  • Outlook calendar is the meeting index (added 2026-07-17). The MS 365 MCP calendar sees every meeting Jeff attended, hosted or not — which the transcript sources cannot guarantee. So before gathering transcripts, pull the calendar for the window to get the authoritative list of what happened (title, time, attendees), then match transcripts against it. This closes the real gap: a meeting Jeff attended but didn't host, with no .vtt dropped, was previously invisible. The calendar surfaces it so Sol can flag the missing transcript rather than silently skip the meeting. The calendar is an index, never a content source — it says a meeting occurred; the transcript is still the only source for what was said.
  • On-demand (org has Claude Code Cloud disabled — no background scheduling; see memory cloud-routines-org-disabled).
  • Propose, don't auto-merge — action-item changes are shown to Jeff for approval before the consolidated list is edited.
  • Audience by meeting type (added 2026-07-28). Most notes are internal working artifacts. But Op Mechs and firm meetings (see _shared/fd-firms.md for the current firm list) are external — Jeff posts these notes to Slack — so they get a humanizer pass and ship without the internal marker. Program-level and FDE-process meetings stay internal. See Step 3.5 for the rule.

When to run this

  • Jeff says "process my meetings," "generate notes from my transcripts," "catch up my action items," or drops .vtt files in team_inbox/meeting_transcripts/.
  • Jeff can scope it: "just today's standup," "since Monday," a specific meeting name.

Step 1 — Resolve scope

Default window = since the last processed meeting:

  1. Find the newest meeting-notes_<YYYY-MM-DD>_*.md in owners_inbox/ (and owners_inbox/archive/). The window runs from that date to now.
  2. Fallback: if none exists, use the previous 48 hours.
  3. If Jeff named a window or a specific meeting, that overrides.

State the resolved window in one line and proceed (don't ask — Jeff will correct if wrong): e.g. "Processing meetings since your last note on 2026-07-09 → now."

Step 2 — Gather transcripts

First build the meeting index from the Outlook calendar (what happened), then gather transcripts to match against it. The drop-folder is primary for content (shared recordings); Zoom is secondary (only meetings Jeff hosts).

2a. Outlook calendar — the meeting index (run first)

Pull the calendar over the resolved window to get the authoritative list of meetings Jeff was on. Load the tool in one call:

ToolSearch select:mcp__claude_ai_Microsoft_365__outlook_calendar_search

Call outlook_calendar_search with query: "*", afterDateTime/beforeDateTime set to the window bounds. Keep only real working meetings: drop all-day items, OOO/holiday blocks, and isCancelled: true events; an event Jeff declined can be dropped too. For each kept event capture subject, start (with its timeZone), and attendees — this is the index, not content. Do not read event bodies as a source of what was said.

The index has two jobs:

  1. It's the checklist transcripts get matched against (by meeting name + date).
  2. Any indexed meeting with no matching transcript (no .vtt, no Zoom asset) is a gap — list it explicitly in Step 2c so Sol can tell Jeff "these happened but have no transcript; export the .vtt if you want notes." Never invent notes for a gap from the calendar alone.

2b. Drop-folder (primary) — .vtt files Jeff exported

Scan team_inbox/meeting_transcripts/ for .vtt files not already represented by a note in owners_inbox/ (match on meeting name + date). These are the main input — most of Jeff's meetings are shared with him, so he exports the transcript and drops it here. Each .vtt is one meeting to process.

2c. Zoom cloud recordings (secondary) — meetings Jeff hosts

Only meetings Jeff hosts appear here; shared-with-him recordings never will (verified 2026-07-10). This step is a bonus fast-path, not the main event. Load the Zoom read tools in one call:

ToolSearch select:mcp__claude_ai_Zoom_for_Claude__search_meetings,mcp__claude_ai_Zoom_for_Claude__get_meeting_assets,mcp__claude_ai_Zoom_for_Claude__recordings_list
  1. search_meetings over the window. The tool requires a timezone — always pass America/Chicago (Jeff is US Central); never stop to ask for it. Set from to the start of the window in UTC.
  2. For each returned meeting with meeting_status: HISTORY, call get_meeting_assets with its meeting_uuid (prefer the UUID over the numeric ID). Pull:
    • my_notes / transcript — the source for the notes.
    • meeting_summary (Zoom's AI recap + next_steps) — use as corroboration only, not as the source of record. Zoom summaries can miss or garble things.
  3. Empty is normal and expected — if search_meetings/recordings_list return nothing (they will, unless Jeff hosted a cloud-recorded meeting in the window), just note it in one line and move on. This is not an error and needs no investigation.

2c-bis. Non-host recordings — the shared retrieval fallback chain

For an indexed meeting Jeff did NOT host with no .vtt in the drop-folder (the common case — the Zoom API in 2c will not surface it), don't jump straight to marking it a gap. First run the shared fallback chain in _shared/zoom-transcript-retrieval.md: it tries the Zoom API, then the share-link token path (find the no-reply@zoom.us "shared a recording with you" email in Outlook, drive the browser (Chrome extension first, Playwright fallback) to the iet=-token share URL, extract the "Audio Transcript" panel), then the Intuit Zoom portal pages. It saves the raw transcript to team_inbox/meeting_transcripts/<meeting-slug>_<YYYY-MM-DD>-transcript.txt — the same drop-folder as the .vtt exports — which then feeds Step 3 like any other transcript and gets filed to the firm repo in Step 6. This save is mandatory: a transcript this skill retrieves must land in that folder before it is turned into notes, exactly as an exported .vtt would — never process transcript text that lives only in the conversation. Follow that file's SSO pause-and-ask-Jeff rule and its fail-loudly failure modes — a meeting only becomes a Step 2d gap after the chain genuinely comes up empty.

2d. Reconcile sources against the index

  • If a meeting appears in both the drop-folder and Zoom, prefer the .vtt (it's the source Jeff curated); treat the Zoom asset as redundant.
  • Match each transcript to an indexed meeting (name + date) to give the note an accurate title and attendee list. A transcript with no matching calendar event is still processed — the index is a helper, not a gate.
  • Gaps: list every indexed meeting that has no transcript from any source — the drop-folder, the Zoom API, and the shared retrieval fallback chain (2c-bis). Only a meeting the chain also failed to retrieve is a true gap. These are reported to Jeff, not processed — never fabricate a meeting or its content from the calendar alone.
  • Never fabricate a meeting or its content. If there are no transcripts at all, report the index + gaps and stop — nothing to process into notes.

List what you're about to process (name + date + source) and the gaps (indexed but no transcript) before dispatching.

Step 3 — Dispatch the notes specialist (one per meeting)

Spawn one subagent per distinct meeting (model: opus — this is synthesis over messy transcript input, matching the slack-recap precedent). The subagent has no session history or memory — the spawn prompt must be fully self-contained. Include all of the following:

Identity + grounding discipline:

You are a meticulous meeting-notes specialist. You never fabricate. Every factual claim traces to the transcript. Anything you infer, or that a speaker stated as uncertain, is tagged [Unverified] inline. A short, honest set of notes is correct — do not pad. End the file with the trace footer specified below.

Who Jeff is (read _shared/fd-firms.md and paste its "Who Jeff is" block verbatim into the spawn prompt — prevents the recurring role/firm errors). That file's identity block already covers the firm list, the Accenture/not-FTE distinction, and the Kaitlyn/Citrin/Op Mech/Rohan Roy glossary. In addition, for this skill specifically: the contracts contact appears across notes as Chung/Chong/Chang — treat as one person and tag the spelling [Unverified].

The transcript for this meeting (paste inline, or give the exact .vtt path to read).

The template to fill (match the existing owners_inbox/meeting-notes_*.md exemplars):

  • A short header block: Team, Meeting Topic, Date & Time, Attendees.
  • ### 1. Executive Summary
  • ### 2. Key Discussion Points (bolded sub-topics)
  • ### 3. Decisions Made
  • ### 4. Action Items — a table with columns Action Item | Owner(s) | Due Date | Priority. Capture every action item, with its owner; if unclear, say "Not Specified."
  • ### 5. Risks / Blockers
  • ### 6. Open Questions / Parking Lot

Output contract for the subagent:

  • Return the full note content to Sol as its result. It also writes the working copy to owners_inbox/meeting-notes_<YYYY-MM-DD>_<slug>.md:
    • First line: the audience marker Sol assigns for this meeting (Sol passes it in the spawn prompt). Internal meetings get <!-- audience: internal -->; Op Mechs and firm meetings get <!-- audience: external --> and will be humanized by Sol in Step 3.5. When in doubt, the subagent uses internal and flags the ambiguity in its report — Sol decides.
    • Last block: the trace footer —
      ---
      **Sources:** [the transcript path or Zoom meeting UUID + the exact date/window]
      **Claims marked [Unverified]:** [list them, or "none"]
  • The subagent does NOT write to Google Drive and does NOT touch the consolidated action-item list. It returns content; Sol routes and reconciles (Steps 4–5). If the firm is ambiguous, the subagent says so in its report rather than guessing.

Dispatch the per-meeting subagents in parallel when there is more than one.

Step 3.5 — Classify audience + humanizer pass (Sol does this)

Do this after the notes come back and before routing them to Drive (Step 3.6), so the version filed to record is the finished one.

Classify each note's audience:

  • External (humanize): the meeting is an Op Mech (Operating Mechanism) OR it routes to a firm folder (see _shared/fd-firms.md for the current firm list). Jeff posts these to Slack, so they must read as human-written.
  • Internal (skip): program-level and FDE-process meetings (the ones that route to Program/00_delivery/). These stay working artifacts.
  • A firm Op Mech is external on both counts — no conflict. If a meeting is genuinely ambiguous (can't tell the firm, or whether it's an Op Mech), treat it as internal and flag it to Jeff rather than guessing.

For each external note, run the humanizer pass (invoke the humanizer skill) as the final editing step, after the content is complete and correct:

  1. Run humanizer over the note body — remove em/en dashes, emojis, rule-of-three, inline-header bullet lists, promotional filler, and the rest, so it reads as Jeff wrote it. Where a voice sample exists (Jeff's prior Slack messages or notes), match it.
  2. Swap the marker: change the first line from <!-- audience: internal --> to <!-- audience: external -->. (The check-humanizer.py hook only skips files marked internal; an external note that still had the internal marker would dodge the very check it needs, and one that has no marker and still contains a dash would be blocked on delivery.)
  3. Keep the trace footer and every [Unverified] marker — humanizing is about tone, not grounding. The Sources / [Unverified] footer stays; factual claims still trace.
  4. Preserve the note's structure (header, the six sections, the action-item table). The humanizer smooths prose; it does not gut the template or drop the action-item table. The table is data, not prose — leave it intact.

Do not humanize internal notes; leave their <!-- audience: internal --> marker in place. This mirrors the standing rule (memory opmech-notes-humanize): Op Mech notes for the firms go to Slack, so they always get humanized, footer and [Unverified] preserved, zero em/en dashes.

Step 3.6 — Route notes to Google Drive (Sol does this)

For each meeting, Sol files a copy to Drive per memory drive-folder-routing. Route the finished note — for external meetings that is the humanized version from Step 3.5.

  • Firm-specific (see _shared/fd-firms.md for the current firm list) → the firm's shared-repo meeting-notes folder <Firm>/00_delivery/meeting_notes/, under the canonical FD Shared Drive (note the trailing space in "Forward Deploy " — always quote the path):
    /Users/jeffheinen/Library/CloudStorage/GoogleDrive-jeff_heinen@intuit.com/Shared drives/MM Partner Program Central Drive/Forward Deploy /<Firm>/
    The firms sit flat directly under that path — NO extra Forward Deploy/ wrapper inside <Firm>/. Test writability first (ls -ld → drwx writable, dr-x blocked); if the folder is absent or read-only, FAIL LOUDLY and leave the note in owners_inbox/ unfiled. Do NOT silently write to a stale path. (This Shared Drive superseded the retired "Agent Studio Forward Deploy" folder on 2026-07-24 — see memory drive-folder-routing.)
  • Non-firm meeting (program-level OR FDE-process — both go to one folder) → Program/00_delivery/meeting_notes/ in the same shared repo. FDE-process is a cross-program topic with no separate notes folder.
  • Unknown / new firm / ambiguous → do NOT guess. Ask Jeff.

See memory drive-folder-routing for the full personal-vs-shared rubric and current layout.

File name: Meeting Notes - <meeting name> - <YYYY-MM-DD>.md. Use the Write tool for the new Drive file (sanctioned per memory permissions-cloud-and-bash). safe-mv is only for moving an existing file on a CloudStorage path (Step 6), never for creating these.

Step 4 — Reconcile action items (propose, do NOT auto-merge)

  1. Collect the Action Items table from every note produced this run.
  2. Read the current consolidated list — the newest owners_inbox/consolidated-action-items_<date>.md.
  3. De-dupe each candidate against both the open-items table and the resolved/closed table (so a closed item does not reappear). Match on substance, not exact wording.
  4. Produce a proposed-changes block for Sol to show Jeff — do not edit the list yet:
    • NEW items — with a suggested priority (🔴/🟡/🟢), due, and source note.
    • UPDATE — candidates that look like a change to an existing row (e.g. new due date).
    • RESOLVED — existing open rows the new notes suggest are now done.
    • Filter to Jeff's items. Items owned by others go under the existing "Gating dependencies" framing only if they touch his firms; Citrin items stay context-only.

Step 5 — Sol review + delivery

  1. Delivery review (per CLAUDE.md): request satisfied, notes match the template, each note has the correct audience marker + trace footer, corrections honored (Kaitlyn; Accenture-not-FTE; the firm list per _shared/fd-firms.md, not Citrin). For external notes (Op Mechs + firm meetings), confirm the Step 3.5 humanizer pass actually ran: marker flipped to external, no em/en dashes or emojis, footer + [Unverified] markers still present, action table intact. Return to the specialist for revision if any note is thin, ungrounded, or misattributes Jeff's role/firms.
  2. Give Jeff a tight chat summary: meetings processed, files written (with Drive folders), and the proposed action-item changes.
  3. Only on Jeff's OK, write a new dated owners_inbox/consolidated-action-items_<today>.md with the approved changes merged in, and archive the prior list to owners_inbox/archive/ via safe-mv.sh (dry-run first). Preserve the list's structure: priority-sorted open table, gating section, resolved/closed table, corrections + [Unverified] footer.
  4. Sync the same approved changes into Apple Reminders — automatically, same approval gate as 3, no second confirmation. Per memory full-sweep-includes-reminders (2026-07-27, Jeff decided action items are managed in the Reminders app, not only the markdown list), the moment Jeff says OK to the action-item diff in step 3, that single approval also covers pushing it into Reminders — this is a low-risk, AUTO write per Jeff's write-autonomy split (CLAUDE.md), not a second propose-then-approve gate.

    Load the tools:

    ToolSearch select:mcp__apple-reminders__create_reminder,mcp__apple-reminders__update_reminder,mcp__apple-reminders__complete_reminder,mcp__apple-reminders__search_reminders
    • NEW items → create_reminder in the Intuit list, name prefixed [FD #N] where N is the item's row position in the freshly-written consolidated list's open table (top to bottom, 1-indexed at write time) — this is the stable numbering convention: N maps to "row N of the open table in the consolidated-action-items file just written," not to any Jira or Binder ID. Set due_date and priority from the row (map 🔴→1/high, 🟡→5/medium, 🟢→9/low).
    • UPDATE items (changed due date or priority) → update_reminder, matched by searching for the existing [FD #N]-prefixed title first (search_reminders) so the edit lands on the right reminder rather than creating a duplicate.
    • RESOLVED items → complete_reminder, matched the same way by [FD #N] prefix.
    • No native tags. Per memory reminders-tags-not-api-settable, the Reminders MCP can only set name / body / due_date / priority — there is no tags field, and appending #hashtag text to a title does not promote to a real tag pill. The [FD #N] prefix in the title is therefore the only available cross-reference mechanism between a reminder and its markdown row — do not attempt to use tags for firm-grouping or anything else.
    • If a [FD #N] numbering collision or ambiguity comes up (e.g. the open table was re-sorted and row numbers shifted), re-derive N from the freshly-written file each run rather than trusting a stale N from a prior run's Reminders titles.
    • Report back to Jeff in the same delivery summary: how many reminders were created, updated, and completed.

Step 6 — Archive inputs and delivered notes

Inputs. team_inbox/meeting_transcripts/ is a transient staging inbox; the copy of record for a raw transcript is now the firm's shared-repo 00_delivery/transcripts/ folder. This applies to every raw transcript that passed through the staging inbox this run — Jeff-exported .vtt files and .txt files this skill retrieved itself via the _shared/zoom-transcript-retrieval.md chain (Step 2c-bis). A retrieved .txt is filed to record exactly like a .vtt; do not treat "I downloaded it" as a reason to skip filing it. The only transcripts that need no filing are ones that never left Zoom (host-owned, read live via the Zoom API and never saved to a file).

  • Firm-specific transcript (.vtt or retrieved .txt): move it from team_inbox/meeting_transcripts/ to that firm's 00_delivery/transcripts/ under the canonical FD Shared Drive (quote the trailing-space path):
    /Users/jeffheinen/Library/CloudStorage/GoogleDrive-jeff_heinen@intuit.com/Shared drives/MM Partner Program Central Drive/Forward Deploy /<Firm>/00_delivery/transcripts/
    Firms sit flat directly under that path (no extra Forward Deploy/ wrapper inside <Firm>/). Use safe-mv.sh (--dry-run first, then real). This is a move of an existing file on a CloudStorage destination, so it goes through safe-mv — never raw mv/cp. If the firm's shared folder isn't mounted, do not archive — leave the file in the staging inbox and report it (same fail-loudly rule as Step 3.6); it can't be filed to record until the folder is mounted.
  • Non-firm / program / FDE-process transcript (.vtt or .txt): move it to the program transcript folder under the same canonical Shared Drive:
    /Users/jeffheinen/Library/CloudStorage/GoogleDrive-jeff_heinen@intuit.com/Shared drives/MM Partner Program Central Drive/Forward Deploy /Program/00_delivery/transcripts/
    via safe-mv.sh (--dry-run first, then real). This is the exact parallel of the firm case — Program/ sits alongside each firm folder (see _shared/fd-firms.md) in the same Shared Drive. Program transcripts have a copy of record too — do not dump them in the local team_inbox/ archive. Same fail-loudly rule: if the folder isn't mounted, leave the file staged and report it.
  • Confirm the inbox is empty of raw transcripts at the end of the run. Every file that came in via Step 2b/2c-bis should be either filed to a firm repo or archived locally — none left loose in team_inbox/meeting_transcripts/ (or team_inbox/ root) unaccounted for. If one is stuck (firm folder unmounted), that's the one allowed exception and it must be reported.
  • Zoom-sourced meetings read live and never saved need no archiving — they live in Zoom.

Working note copies. For each note produced this run, once its Drive copy is confirmed written (Step 3.6), move the owners_inbox/meeting-notes_<date>_<slug>.md working copy to owners_inbox/archive/ via safe-mv.sh (--dry-run first, then real). The Drive copy is the copy of record; owners_inbox/ is a delivery surface, not storage, per the inbox-lifecycle rule in CLAUDE.md. Never archive a note whose Drive copy is not confirmed — if Drive routing was skipped or blocked (e.g. an ambiguous firm you asked Jeff about), leave that note in the inbox root until it is filed. Do not archive the consolidated action-item list here (it is the live master, archived only when superseded in Step 5.3).

Notes

  • One meeting = one note = one subagent. Don't merge two meetings into one note.
  • Sol owns Drive routing and the list edit; the subagent only produces note content. This keeps the firm-ambiguity rule and the propose-then-approve gate under Sol's control.
  • Grounding is non-negotiable: no footer / no [Unverified] markers on a note that makes factual claims = not finished. The check-trace-footer.py and check-humanizer.py hooks enforce the output side; the internal marker satisfies the humanizer hook.
  • Non-host transcripts use the shared fallback chain in _shared/zoom-transcript-retrieval.md (Zoom API → share-link token via the browser (Chrome extension first, Playwright fallback) → portal pages). It is the single source of truth for retrieval; this skill only calls it and consumes the .txt it produces.
  • Related: _shared/fd-firms.md (firm list + identity block), memories drive-folder-routing, jeff-role-forward-deploy, slack-recap-on-demand, permissions-cloud-and-bash, intuit-tooling-stack (Outlook = email + calendar transport via the MS 365 MCP), full-sweep-includes-reminders (Apple Reminders sync, Step 5.4), reminders-tags-not-api-settable ([FD #N] prefix, no native tags).

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.