Secret Scanning Alert Investigator

Critical Requirements

Evidence-Based Analysis Only

DO NOT make assumptions or guesses. All reported information must be:

• Validated with actual API responses, git history, or file contents
• Cited with specific commit SHAs, timestamps, or API endpoints
• Clearly distinguished from estimates or possibilities

When providing estimates or possibilities (e.g., in risk assessment):

• Explicitly label them as "ESTIMATE" or "POSSIBLE"
• Explain the logic and evidence used to arrive at them
• State what evidence would be needed to confirm

Modification Safeguards

Any action that modifies state requires explicit safeguards:

1. Warn the user about the specific action and its consequences
2. Request explicit confirmation with "CONFIRM" before proceeding
3. Never assume the user wants to resolve, close, or modify anything

Actions requiring double-confirmation:

• Resolving or closing alerts
• Adding comments to alerts
• Any GitHub API PATCH/POST/DELETE operations
• Creating issues, PRs, or gists
• Running active verification against third-party services
• Executing history cleanup commands
• Installing tools on the user's system

Permitted without confirmation:

• GET/read operations
• Creating/deleting local temporary clones in /tmp
• Writing reports to user-specified local paths
• Passive analysis of already-retrieved data

Parallel Execution Architecture

This skill uses parallel sub-agents to accelerate investigation:

Phase 0: Tool Check (sequential)
    |
Phase 0.5: Permission Pre-Approval (sequential - user interaction)
    |
Phase 1: Mode Selection (sequential - user interaction)
    |
+-----------------------------------------------------------+
|                    PARALLEL BLOCK A                        |
|  Agent A1: Alert Acquisition    -> Full alert + locations  |
|  Agent A2: Repository Clone     -> Clone + file history    |
|  Agent A3: Commit/PR Context    -> Author, PR, issues      |
+-----------------------------------------------------------+
    | (wait for all)
+-----------------------------------------------------------+
|                    PARALLEL BLOCK B                        |
|  Agent B1: TruffleHog Scan      -> Detection + verify      |
|  Agent B2: Gitleaks Scan        -> Cross-validate + SARIF  |
|  Agent B3: Entropy Analysis     -> Shannon entropy         |
|  Agent B4: Service Analysis     -> GitHub App / token      |
+-----------------------------------------------------------+
    | (wait for all)
Phase 6.2: Active Verification (requires confirmation)
    |
Phase 7: Risk Assessment (synthesis)
    |
Phase 8: Report & Actions (user interaction)

Sub-Agent Invocation

Use the Task tool to launch parallel agents:

# Launch Parallel Block A (all three in single message)
<Task tool call 1: subagent_type="Explore", prompt="...", run_in_background=true>
<Task tool call 2: subagent_type="Explore", prompt="...", run_in_background=true>
<Task tool call 3: subagent_type="Explore", prompt="...", run_in_background=true>

READ: references/sub-agents/block-a.md for Agent A1-A3 prompts. READ: references/sub-agents/block-b.md for Agent B1-B4 prompts and coordinator.

Artifact Storage Pattern

All agents write findings to /tmp/secret-scan-{alert_number}/:

alert.json       (Agent A1)
provenance.json  (Agent A2)
context.json     (Agent A3)
trufflehog.json  (Agent B1)
gitleaks.json    (Agent B2)
entropy.json     (Agent B3)
service.json     (Agent B4)

The coordinator reads all artifacts ONCE during synthesis phase.

Phase Overview

Phase 0: Environment Setup

Verify required tools (gh, git) and offer to install optional tools (trufflehog, gitleaks, ggshield).

READ: references/phases/phase-0-setup.md for tool check details and installation commands.

Phase 0.5: Permission Pre-Approval

Present all read-only commands upfront for batch approval to minimize interruptions.

READ: references/phases/phase-0.5-permissions.md for permission request format.

Phase 1: Initial Triage and Mode Selection

Fetch basic alert details and require user to select PASSIVE or ACTIVE mode.

READ: references/phases/phase-1-triage.md for mode selection prompts.

Phases 2-6: Investigation

Deep investigation including alert acquisition, multi-tool scanning, provenance tracing, and secret-specific analysis.

READ: references/phases/phases-2-6-investigation.md for detailed procedures.

Phase 7: Risk Assessment

Evaluate risk factors and produce overall rating (CRITICAL/HIGH/MEDIUM/LOW).

READ: references/phases/phase-7-risk.md for risk criteria and assessment format.

Phase 8: Report Generation and Actions

8.1 Generate Report

After investigation completes, generate comprehensive report.

READ: references/report-template.md for the full report structure.

8.2 Output Options Menu

===============================================================
                    INVESTIGATION COMPLETE
===============================================================

Investigation Mode: {ACTIVE/PASSIVE}
Alerts Analyzed: {count}
Risk Rating: {CRITICAL/HIGH/MEDIUM/LOW}

OUTPUT OPTIONS:
1. Display report here
2. Save report to local file
3. Display and save

ADDITIONAL ACTIONS AVAILABLE:
4. Resolve alert(s) on GitHub
5. Generate SARIF output for security dashboard
6. Generate history cleanup commands
7. Generate prevention recommendations (pre-commit hooks, CI/CD)
8. Check for public leaks (requires ggshield)
9. Create GitHub issue with findings
10. Export timeline as CSV
11. Generate permanent permissions (settings.json)

Which options would you like? (comma-separated, e.g., "1,4,6"):

Action Reference Files

Each action has detailed implementation guidance:

READ: references/actions/rotation-guides.md for credential rotation instructions.

Batch Processing

When multiple alerts are provided:

1. Run Phase 0 (tool check) once
2. Ask for mode selection once (applies to all unless user specifies otherwise)
3. Process each alert through Phases 2-7
4. Generate combined report with:
   • Shared executive summary noting patterns
   • Individual alert sections
   • Combined timeline showing relationships
   • Aggregate risk assessment
5. For Phase 8 actions, offer batch operations with explicit per-alert confirmation

Error Handling

Tool Requirements Summary

All optional tools are free and open-source.

Example Invocations

# Single alert by URL
"Investigate https://github.com/org/repo/security/secret-scanning/19"

# Single alert by components
"Investigate secret scanning alert #19 in org/repo"

# Multiple alerts
"Investigate secret scanning alerts #19, #20, #21 in org/repo"

# All open alerts in repo
"Investigate all open secret scanning alerts in org/repo"

# Organization-wide
"Investigate all critical secret scanning alerts in the org organization"