output

Presentation format for all query results. MUST be used after every ad-hoc query.

SKILL.md

GitHub

Output Skill

Defines the user-facing output format for all queries — Inventory, External Probe, Policy & Config, Compound, and general questions.


Two-Phase Output

Every response has two phases:

  1. Processing phase — CLI commands and their full output are shown inline as they run (mandatory per CLAUDE.md). This is the audit trail.
  2. Results phase — After all processing, insert --- then ## Results. This is the clean answer the user reads.

The --- / ## Results separator is mandatory. Everything above it is working output; everything below it is the structured answer.

Pipeline Visibility

Pipeline internals (Claim Envelopes, Expert Responses, Critic Responses) are never shown proactively — they run silently.

What IS always shown: Every aws, curl, nmap, testssl, pmapper command and its full output — see CLAUDE.md → CLI format rule.

On request only: If the user asks ("show me the claim envelope", "why was this scored 4?"), show the relevant pipeline detail inline.

Complete Enumeration Rule

Every affected resource MUST be listed in full. NO EXCEPTIONS.

Never use: "...and 30 more", "(truncated)", "Multiple resources affected", "Various policies including...", "etc.", "and others", "..." to indicate more items.

If a finding affects 30 policies, list all 30 with full ARNs.


Section Activation

Section Inventory External Probe Policy & Config General Question
Header Always Always Always Always
Scope Always Always Always Always
Findings If issues Always Always Always
Impact If issues Always Always Always
Attack Chains — If chained If chained If applicable
Blast Radius — If chained If applicable If applicable
Cross-Account Summary Multi-account only Multi-account only Multi-account only Multi-account only
Recommendations If issues Always Always Always
Confidence Always Always Always Always

Rule: Omit inactive sections entirely — no empty headers. Clean Inventory with no issues: Header + Scope + Confidence only.


Section Templates

Header

Single-account mode:

---
## Results

### {{Short title — what was asked and answered}}
{{Inventory | External Probe | Policy & Config | Compound}} | {{YYYY-MM-DD}} | {{account_id}} | SCP: {{AVAILABLE | USER_PROVIDED | UNKNOWN}}

Multi-account mode (when MULTI_ACCOUNT_STATUS = AVAILABLE):

---
## Results

### {{Short title — what was asked and answered}}
{{Inventory | External Probe | Policy & Config | Compound}} | {{YYYY-MM-DD}} | {{N}} accounts | SCP: per-account

In multi-account mode, per-account results use sub-headers:

#### Account: {{account_name}} ({{account_id}}) | SCP: {{status}}

Scope

**Scope:** {{resources checked}} | **Regions:** {{regions}} | **Gaps:** {{what wasn't checked and why}}
**Suppressions:** {{None | list with action taken}}

Findings

#### F{{N}}: {{short title}} · `{{SEVERITY}}` · `{{CONFIRMED|ERROR}} {{N}}/5`
**Resource:** `{{ARN}}` | **Claim:** `{{claim_id}}`
**Account:** `{{account_id}}` (`{{account_name}}`)  ← only in multi-account mode

{{One paragraph — what is misconfigured, what it enables, what guardrail is missing.}}

Rules:

  • For Inventory with no findings: No security findings. {{N}} resources enumerated.
  • FALSE_POSITIVE: omit entirely
  • SUPPRESSED: mention in Scope only
  • Finding numbering (F1, F2...) is local to the response
  • Complete enumeration applies — see CLAUDE.md

Impact

### Impact
**Data at risk:** {{specific types — customer PII, credentials, application data, none}}
**Services at risk:** {{what could be disrupted or manipulated}}
**Exposure:** {{Immediate | Conditional (requires X) | Latent (requires future change)}}
**Compliance:** {{CIS, SOC2, PCI-DSS, GDPR | N/A}}
**Amplifying:** {{no boundary, wildcarded resource, public exposure, no MFA}}
**Mitigating:** {{no instances attached, deny policies, scoped resource}}
**Unknown:** {{SCP status, network controls, other accounts}}

Frame from the defender's perspective. This is the section a manager reads.

Attack Chains

Use the mandatory Attack Path format from agents/roles.md → Exploit Chain Analysis. Substitute F{{N}} Finding references for Claim IDs, and add **Exploitability:** {{IMMEDIATELY EXPLOITABLE | REQUIRES PIVOT | LATENT RISK}}.

Blast Radius

### Blast Radius
| Resource Type | Scope | Access Level | Condition |
|:-------------|:------|:-------------|:----------|
| {{type}} | {{how many / which}} | {{Read/Write/Delete/Admin}} | {{what must be true}} |

Only include when the query involves compromise scenarios or chains reveal lateral movement. Complete enumeration applies — see Complete Enumeration Rule above. For full identity-level blast radius analysis (dedicated "what if X is compromised?" queries), use skills/identity-blast-radius/SKILL.md instead — it has its own detailed output format.

Cross-Account Summary (multi-account mode only)

### Cross-Account Summary
| Account | ID | Findings | Critical | High | Medium | Low | Status |
|:--------|:---|:---------|:---------|:-----|:-------|:----|:-------|
| {{name}} | {{account_id}} | {{total}} | {{count}} | {{count}} | {{count}} | {{count}} | {{Complete | Skipped — reason}} |

Rules:

  • Include this section after all per-account findings and before Recommendations
  • Every configured account appears — including skipped/errored ones
  • Severity counts only include Score 4-5 confirmed findings
  • This table provides cross-account visibility but does NOT analyze relationships between accounts (Tier 1 limitation)

Recommendations

### Recommendations
| Priority | Action | Effort | Risk Reduction |
|:---------|:-------|:-------|:---------------|
| P0 | {{immediate action}} | Low/Med/High | {{what it eliminates}} |
| P1 | {{short-term action}} | Low/Med/High | {{what it eliminates}} |

Order by risk reduction impact. Include CLI commands for actionable items. Effort: Low (<1 hour config change), Med (policy rewrite, testing), High (architecture change, multi-team).

Confidence

### Confidence: {{N}}/5
{{What couldn't be verified}} | **Verdict:** {{Accept as-is | Verify X manually | Needs further investigation}}

For pipeline claims, use the Critic's score (lowest if multiple). For Inventory/External Probes (no Critic): 5 = direct evidence no gaps, 4 = strong with minor gaps, 3 = needs manual verification.


Compact Mode

For single-finding queries with no chains or blast radius (e.g., "is encryption enabled on bucket X?"):

---
## Results

### {{Title}}
{{type}} | {{date}} | {{account_id}}

**Scope:** {{one line}}

#### {{short title}} · `{{SEVERITY}}` · `{{STATUS}} {{N}}/5`
**Resource:** `{{ARN}}`

{{Issue paragraph}}

**Recommendation:** {{single action with CLI command}}

**Confidence:** {{N}}/5 — {{one-line verdict}}

When in doubt, use the full format.