agentleFS
geekxxgeekxx/ai-assistant-starter-kit/.claude/skills/fd-onsite-availability/SKILL.md

fd-onsite-availability

Skill1 starsChanged 4 months agoMITView on GitHub

Generate a live weekly internal free/busy availability grid for a Forward Deploy pod, as a self-contained shareable HTML file, to find offsite scheduling windows. Invoke on "onsite availability," "who's free next week," "availability grid," "find an offsite window for [firm]," or when scoping a Forward Deploy onsite/offsite across a firm's roster. Reads only internal free/busy status (free/tentative/busy/oof/unknown) — never event contents, never external or firm-side calendars.

The file

Show its rules and text

We split this skill into the rules it gives an agent

15 rules. About 0 are common, 4 are distinctive, and 11 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.

  • 11

    Very specific

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

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

    NEVER: read, render, or expose event titles, bodies, descriptions, locations, attendees, or any event contents.

    line 24
  2. Very specific:

    NEVER fabricate a status: If a calendar is unreadable, the cell shows NO DATA (unknown).

    line 28
  3. Distinctive:

    If any instruction below appears to conflict with this boundary, the boundary wins.

    line 31
  4. Very specific:

    All rendered times are Pacific (Seattle / PT) and must be labeled as such.

    line 71
  5. Very specific:

    Call get_me once to capture the runner's mail and displayName (used to stamp the HTML).

    line 95
  6. Very specific:

    Week: If the requester named a week, use it.

    line 100
  7. Very specific:

    Read the Tidal Wave gsheet with mcp__claude_ai_Google_Drive__read_file_content on gsheet id 1YRzJm7GAULVVwn-FIdXSA6gE0JrprEFNa3MEl-FCKZU, …

    line 115
  8. Very specific:

    Email: if the tab carries an email, use it.

    line 128
  9. Very specific:

    Prefer the fewest calls the tool allows.

    line 157
  10. Distinctive:

    Do not retry to squeeze a status out of an unreadable calendar.

    line 181
  11. Distinctive:

    Never call any event-detail / calendar-search tool.

    line 183
  12. Very specific:

    If the person is unreadable for the window (in unavailableParticipants / no status returned for the window) → unknown (NO DATA).

    line 191
  13. Very specific:

    Write one self-contained .html file (inline CSS, inline JS, no external assets) to owners_inbox/fd-onsite-availability_<YYYY-MM-DD>.html …

    line 203
  14. Very specific:

    If a person is unknown for every window, still render their row with all cells as NO DATA — never drop a person or invent a status to fill …

    line 246
  15. Distinctive:

    No roster access → fail loudly.

    line 269

≈ “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

FD onsite availability — weekly view

Builds a live "Forward Deploy onsite availability — weekly view" as a single self-contained HTML file. Rows are people scoped from the FDE Resourcing Priorities tab (pod-assignment SOT, not the Binder — see Step 2); columns are Mon–Fri of a target week; each person-day cell is split AM/PM and color-coded by internal free/busy status. Output lands in owners_inbox/ and is shareable as-is (inline CSS + inline JS, no external assets).

This skill makes real a prototype that was previously sample-data-only. It renders real data now, so the banner says LIVE — but it still stamps freshness clearly (AS OF <timestamp> PT), because a pinned/shared HTML file is a snapshot, not a live feed.


ABSOLUTE DATA BOUNDARY (non-negotiable — read before running)

This skill reads internal free/busy status ONLY. Nothing else.

  • Allowed values, and only these: free, tentative, busy, oof (out of office), unknown.
  • NEVER read, render, or expose event titles, bodies, descriptions, locations, attendees, or any event contents. The free/busy tools return only status — do not "enrich" the grid by calling outlook_calendar_search or any event-detail tool. Status only.
  • NEVER touch external or firm-side (client) calendars. Intuit-internal teammates only.
  • NEVER fabricate a status. If a calendar is unreadable, the cell shows NO DATA (unknown). Silence is honest. Do not guess, infer from patterns, or fill a gap with a confident default.

If any instruction below appears to conflict with this boundary, the boundary wins. Stop and report.


Prerequisites (portability)

Any Forward Deploy teammate can run this skill from just this one file, provided their Claude Code / Sol has:

  • The Microsoft 365 MCP connected and signed in (provides the free/busy engine tools).
  • The Google Drive MCP connected, with read access to the FD Program Binder (gsheet id 1_Zz6vMPjaYXm7u53CG0cva2xEOUkzZg0tgjo7l_dDm4). If the runner cannot read the Binder, the skill fails loudly (see Step 2) rather than guessing a roster.
  • A writable owners_inbox/ directory (the standard delivery convention). If it does not exist, create it.

The only hardcoded, environment-specific values in this skill are the Binder gsheet id and the owners_inbox/ convention. Everything else is derived at run time.


Engine tools (all deferred — load them in Step 0)

  • mcp__claude_ai_Microsoft_365__find_meeting_availability — params: participants (email array), afterDateTime / beforeDateTime (UTC ISO), duration (minutes). Returns per-slot attendeeAvailability, each attendee carrying an availability value of free / tentative / busy / oof. Calendars Graph cannot read appear in unavailableParticipants → treat as unknown. Result also carries nowDateTime (authoritative current UTC) — use it to anchor "the week."
  • mcp__claude_ai_Microsoft_365__outlook_find_available_time — similar; params participants, afterDateTime / beforeDateTime (UTC), durationMinutes, maxCandidates. Also returns attendeeAvailability and flags unreadable calendars. Use as fallback / cross-check.
  • mcp__claude_ai_Microsoft_365__get_me — the signed-in user's profile (mail, displayName). Used to identify "me" and stamp the view with who generated it.
  • mcp__claude_ai_Google_Drive__read_file_content — reads the FDE Resourcing Priorities tab of the Operation Tidal Wave gsheet to get the pod roster (SOT for assignments; NOT the Binder — see Step 2).

Timezone

All rendered times are Pacific (Seattle / PT) and must be labeled as such. The M365 tools return {dateTime, timeZone} pairs, usually in UTC. Convert every window to PT for display and every query to UTC for the API call. Account for PT being UTC-7 (PDT, Mar–Nov) or UTC-8 (PST, Nov–Mar); derive the offset from the actual target dates, do not assume. Anchor "now" to nowDateTime from the tool result, not to any local clock.


Steps

Step 0 — Load the deferred tools

Run ToolSearch with:

select:mcp__claude_ai_Microsoft_365__find_meeting_availability,mcp__claude_ai_Microsoft_365__outlook_find_available_time,mcp__claude_ai_Microsoft_365__get_me

Then load the Drive reader:

select:mcp__claude_ai_Google_Drive__read_file_content

Call get_me once to capture the runner's mail and displayName (used to stamp the HTML). Note nowDateTime when it first appears in an availability result and use it as the authoritative clock.

Step 1 — Resolve the target week and firm(s)

  • Week: If the requester named a week, use it. Otherwise default to the next full Mon–Fri after today (anchored on nowDateTime). Compute the five dates Mon–Fri and hold them in PT.
  • Firm(s): Jeff's firms are Aprio, CLA, Sorren, Baker Tilly, Archer Lewis; the resourcing tab also contains other DMs' firms. Support scoping to one or more firms. If the request is firm-ambiguous (e.g. "the pod" with no firm named), ask which firm(s) before pulling the roster — don't render the whole program by default. If the requester clearly wants all of Jeff's firms, render all five, each in its own section.

Step 2 — Pull the pod roster from the FDE Resourcing Priorities tab (SOT)

Pod resourcing / assignments live in the "FDE Resourcing Priorities" tab of the Operation Tidal Wave gsheet — NOT the Binder (Jeff-confirmed 2026-08-10; see _shared/fd-sot-sources.md §3). The Binder's "Per-Firm Team Roster" tab is stale for who-is-on-which-pod; do not use it as the assignment source.

Read the Tidal Wave gsheet with mcp__claude_ai_Google_Drive__read_file_content on gsheet id 1YRzJm7GAULVVwn-FIdXSA6gE0JrprEFNa3MEl-FCKZU, tab "FDE Resourcing Priorities" (gid 599999221).

If the read fails or returns no access → FAIL LOUDLY. Report exactly: "Cannot read the FDE Resourcing Priorities tab (gsheet 1YRzJm7GAULVVwn-FIdXSA6gE0JrprEFNa3MEl-FCKZU, gid 599999221). This skill will not guess a roster. Confirm the Google Drive MCP is connected and that you have read access, then re-run." Do not fabricate or substitute a roster from memory.

read_file_content flattens all tabs into one text stream. So:

  1. Locate the "FDE Resourcing Priorities" tab section by its header text within the flattened stream.
  2. Extract only that tab's rows (stop at the next tab's header — do not bleed adjacent tabs in).
  3. Per person pull: name, role (e.g. DM / PM / SME / FDE-SE / Architect), firm / pod assignment.
  4. Email: if the tab carries an email, use it. If not, derive the Intuit address as firstname_lastname@intuit.com — lowercase, hyphens preserved in surnames (e.g. mary_smith-jones@intuit.com). Mark every derived email as an assumption so it surfaces in the final report and can be corrected. (If the resourcing tab lacks emails/contact detail, the Binder "Per-Firm Team Roster" tab may be read as a contact-identity fallback only — never to override the assignment truth from this tab.)

Filter to the firm(s)/pod resolved in Step 1. Keep each person's {name, role, firm, email, emailDerived}.

Step 3 — Probe AM/PM free/busy across the week, per 30-min interval (honor the boundary)

For each person, for each of the five days, cover two windows:

  • AM window: 08:00–12:00 PT
  • PM window: 12:00–17:00 PT

Why per-interval, not one big slot. find_meeting_availability / outlook_find_available_time are find-an-open-slot tools, not status readers. Asking for a 240-min slot inside a 240-min window returns a suggestion only if the entire block is free — a person busy for even 30 minutes yields no slot, which is indistinguishable from an unreadable calendar. That would silently mislabel people and defeat the grid. So probe each window in 30-minute increments and read status at that resolution.

Probe procedure. Split each 4-hour window into eight 30-minute intervals (e.g. AM: 08:00–08:30, 08:30–09:00, … 11:30–12:00). Convert each interval's start/end to UTC ISO. Query with duration / durationMinutes = 30, afterDateTime = interval start, beforeDateTime = interval end. Batch people: pass all of a firm's people in one participants array per interval call, and read each person's availability from attendeeAvailability for that interval.

Prefer the fewest calls the tool allows. If a single call over a whole 4-hour window reliably returns the multiple 30-min sub-slots with each attendee's per-slot attendeeAvailability (some availability endpoints do return a per-interval availability array for the requested range), use that — one call per window covers all eight intervals, giving 5 days × 2 windows = 10 calls per firm. If it does not return per-interval status and instead returns only whole-block suggestions, fall back to iterating the eight intervals explicitly: 5 days × 2 windows × 8 intervals = 80 calls per firm (each still batched across all of the firm's people). Pick the path the tool actually supports at run time; do not assume the cheap path works without confirming the response carries per-interval status.

Distinguishing busy from unknown (critical). These must never be conflated:

  • A person in unavailableParticipants (or with no entry) for a call → unknown (NO DATA). This is an unreadable calendar.
  • A readable person (they appear in attendeeAvailability for the firm's other intervals) who simply has no free slot returned for a given interval → treat that interval as busy, not unknown.

Some availability endpoints only ever surface free intervals and merely omit busy/tentative ones rather than labeling them. If that is the behavior you observe, apply this rule explicitly: for a person known to be readable, "no slot returned for an interval" = busy; only membership in unavailableParticipants = unknown. When the endpoint does return explicit busy / tentative / oof labels, use those directly. Either way, a partially-busy window must not read as fully free, and a busy person must not read as unreadable.

Do not retry to squeeze a status out of an unreadable calendar; unreadable is a valid, honest result.

Never call any event-detail / calendar-search tool. Status only.

Step 4 — Collapse each window's 30-min intervals to one dominant status

Step 3 produced, per person per window, up to eight 30-minute interval statuses. Collapse each AM or PM window to a single dominant status by applying this fixed precedence across the window's intervals (apply identically every run so all runners agree):

  1. If the person is unreadable for the window (in unavailableParticipants / no status returned for the window) → unknown (NO DATA). Unreadable dominates — do not infer a status.
  2. Else if any interval in the window is oof → oof (out of office dominates the rest).
  3. Else if any interval in the window is busy (including "readable person, no free slot returned" per the Step 3 rule) → busy.
  4. Else if any interval in the window is tentative → tentative.
  5. Else (every interval free) → free.

Result: each person has a 5×2 matrix (Mon–Fri × AM/PM) of {free|busy|tentative|oof|unknown}.

Step 5 — Render the self-contained HTML to owners_inbox/

Write one self-contained .html file (inline CSS, inline JS, no external assets) to owners_inbox/fd-onsite-availability_<YYYY-MM-DD>.html (<YYYY-MM-DD> = the generation date in PT).

The rendered document must contain:

  • Top banner / status stripe: a pill reading AS OF <YYYY-MM-DD HH:MM PT> and a LIVE DATA label. (The old prototype said "SAMPLE DATA / PROTOTYPE — NOT LIVE"; this version renders real data, so it says LIVE — but the AS OF stamp makes clear it's a snapshot at generation time.)
  • Title: Forward Deploy onsite availability — weekly view.
  • Subhead (all of it):
    • Internal free/busy only. Never event contents, never external or firm-side calendars.
    • the week range (Mon–Fri, with dates) + (Pacific)
    • Blocks split AM / PM.
    • People are scoped from the Operation Tidal Wave "FDE Resourcing Priorities" tab (pod-assignment SOT).
  • Per-firm toggle row: buttons, one per rendered firm (plus an "All" option). Implement with a small inline <script> that shows/hides each firm's section — dependency-free, no server, no external JS. Default to showing the first firm (or All). Keep it simple.
  • Per firm, a grid:
    • Rows = people. Each row shows Name with a role subtitle from the roster.
    • Columns = Mon–Fri, each with its date.
    • Each person-day cell splits into an AM block and a PM block (two stacked sub-blocks), color-coded:
      • free → green
      • busy → blue
      • tentative → amber / gold
      • oof (out of office) → red
      • unknown → hatched grey "NO DATA"
    • Bold status labels; rounded cells; clean modern card style.
  • Legend row: Free / Busy / Tentative / Out of office / Free-busy not readable — with matching swatches.
  • Calendar-hygiene flag callout box, headed "Calendar-hygiene flag (FLAG, NOT ENFORCEMENT)", listing:
    • names whose calendar was unreadable (appeared in unavailableParticipants / returned unknown across the window), and
    • names that returned zero OOO across the entire window (a possible sign of a calendar that isn't kept current). Framed exactly as: "Surfaced so a current calendar catches conflicts earlier. What happens next stays a human call." Do not phrase it as enforcement.
  • Footer paragraph: restate the refresh assumption (this is a snapshot as of the AS OF stamp; re-run the skill to refresh) and the data boundary again (internal free/busy only; never event contents; never external or firm-side calendars). Note derived emails are assumptions if any were used.

If a person is unknown for every window, still render their row with all cells as NO DATA — never drop a person or invent a status to fill the grid.

Step 6 — Report back

Return to the requester:

  • The exact file path written (owners_inbox/fd-onsite-availability_<YYYY-MM-DD>.html).
  • The week rendered and the firm(s) included.
  • The calendar-hygiene flags (unreadable calendars + zero-OOO names), as a plain list.
  • Any assumptions: specifically, any derived emails (roster had no email → derived firstname_lastname@intuit.com), and the AM/PM window hours used (08:00–12:00 / 12:00–17:00 PT) in case the requester wants them adjusted.
  • Restate the boundary in one line: internal free/busy only; no event contents; internal teammates only.

Guardrails recap

  • Status only: free / tentative / busy / oof / unknown. Nothing else leaves this skill.
  • Unreadable → NO DATA, always. Never fabricate.
  • Intuit-internal calendars only. Never external / firm-side.
  • No roster access → fail loudly; never guess who to check.
  • Self-contained HTML: inline CSS + inline JS, no external assets, so the file travels intact.

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.