summit53-external-skillsSAFE

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

platforms/claude/skills/  Claude packages
platforms/codex/skills/   Codex packages, including agents/openai.yaml
shared/canonical/         Shared preflight, evidence and asynchronous-lifecycle contracts
manifests/                Deterministic package fingerprints
tools/                    Validation, synchronization and fingerprint tools

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: "ae-deal-inspection-prep"
description: "Prepare the owning account executive for an internal inspection of one named Summit53 opportunity: a concise, read-only one-page evidence sheet with current deal facts, truthful gaps, likely manager questions, defensible answer evidence, and rep-owned next actions. Use when an AE says 'prepare me for deal inspection', 'prep me for my deal inspection on [deal]', 'help me answer questions on this deal', 'what am I missing before my 1:1', 'give me an inspection cheat sheet', or 'what will my manager ask about [deal]'. Do not use for a manager-led formal deal review deck — use deal-review-prep. Do not use for a contested chronology or 'when did it go dark' — use deal-timeline-defence. Do not use for drafting re-engagement outreach — use deal-action-engine, if installed."
version: "1.1.0"
baselined: "2026-10-13"
status: "active"
---

# AE Deal Inspection Prep

Prepare the owning AE to answer from evidence without pretending that missing qualification,
stakeholder, activity, or next-step data exists. Keep the workflow read-only and concise.

## Route first

Route before reading any data:

| Request | Route |
| --- | --- |
| AE-owned readiness for one named deal: likely questions, evidence gaps, answers, next actions | Use this skill |
| Manager-led formal deal review, pipeline-review deck, broad coaching assessment, or executive artifact | Route to `deal-review-prep` |
| Challenged chronology, disputed activity sequence, "when did it go dark", or rescue-decision defence | Route to `deal-timeline-defence` |
| Draft or send re-engagement outreach, write a CRM note, or execute an action | Route to `deal-action-engine`, if installed; do not perform it here |

Persona language alone is insufficient. A manager asking for an AE's formal review deck still
belongs to `deal-review-prep`; an AE challenging the chronology still belongs to
`deal-timeline-defence`.

## Preflight

Read [references/connector-preflight.md](references/connector-preflight.md) and complete its
checklist before any data call. It covers connector and tenant resolution
(`current_context` first, `scoped_org_id`), loading deferred tools in one ToolSearch batch,
`list_available_tools` and live-schema inspection, read-only discipline, fixture exclusion,
unavailable-not-zero, correlational language, and delivery.

Skill-specific gates, stricter than the preflight:

1. Stop before any opportunity lookup if tenant scope is missing or mismatched.
2. Resolve exactly one opportunity by ID or an unambiguous search. Confirm the requester is
   the owning AE when ownership materially affects access or the framing of actions.
3. Require a non-empty `get_opportunity` payload whose `id` and opportunity citation match
   the selected ID.
4. Compare `current_context.user.id` with the opportunity `owner_id`. When they differ,
   require explicit on-behalf-of framing, or route a manager-led request to
   `deal-review-prep`.
5. Read-only: never create notes, update CRM fields, draft or send messages, create
   interventions, or invoke approval-gated actions.
6. Demo label (commercial use): when the connector or `current_context` identifies a demo,
   test, or staging deployment and provenance is not reliable on every row, label every
   output exactly **Demo data — not for customer or commercial decisions.**

## Gather the minimum evidence

Use logical capabilities only when exposed and permitted. Inspect each live schema first.

| Capability | Purpose |
| --- | --- |
| `current_context` | Confirm current organization and viewer context |
| `search_opportunities` | Resolve a supplied deal or account name when no ID is given |
| `get_opportunity` | Current deal, owner, stage, amount, close date, category, and freshness |
| `get_opportunity_notes_summary` | Bounded deal narrative; fall back to recent `get_opportunity_notes` |
| `get_opportunity_activities` | Recent cadence, last observed touch, and recorded next-step evidence |
| `list_methodology_playbooks` | Resolve the published framework definition; ask the user if more than one active framework remains |
| `opportunity_framework_summary` | Qualification evidence for the selected supported framework; do not average absent elements |
| `get_opportunity_signals` | Access-gated ranked constraint and movement; optional, never a blocker |
| `opportunity_stakeholder_intelligence` | Buying-committee coverage and explicit gaps; optional |
| `search_target_accounts`, `get_target_account` | Optional, Outreach plan: committee coverage from the converted outreach target account (see below) |
| `risk_heatmap` | Optional portfolio response: page and select the exact opportunity ID locally, disclose coverage, and treat bounded-page absence as unavailable |
| `meeting_capture_history` | Optional, access-gated: recorded calls for this deal (filter by `opportunity_id`; default `view` is the caller's own sessions) |
| `meeting_context` | Optional, access-gated: bounded intelligence for one recorded session by `session_id`; request a transcript only when a specific question needs it, and keep it within the capped segment and character limits |

Retrieve bounded results sufficient for the inspection horizon. Disclose truncation. Do not
pull a broad account or team review merely to fill the page.

**Recorded call evidence (optional).** When `meeting_capture_history` is exposed and permitted,
list sessions for the exact `opportunity_id` within the inspection horizon, then read at most
the two or three most relevant sessions with `meeting_context`. Use them to corroborate or
contradict what the notes say a buyer committed to. Cite each as session ID plus start time.
Treat meeting intelligence as observed only for what the session record states; model-produced
summaries inside it are still **Inferred**. If the capability is unavailable, denied, or returns
no session for the deal, record "recorded call evidence unavailable" in Sources and limits —
never "no calls happened".

**

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (1)

Gates applied: no_behavioural_pass, no_license.