The Agent Skills Directory

[SAFE]: The skill is primarily instructional and focused on document generation. It does not perform network requests, download external scripts, or execute dangerous shell commands. All operations are confined to reading authorized project documentation and writing a specific specification file.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection because its core function involves processing external incident logs and trace data.
Ingestion points: The skill processes incident evidence, including trace bundles, tool logs, and customer-affected lists, as described in SKILL.md and logic.prompt.
Boundary markers: The workflow includes decision rules that mandate validating provenance, freshness, and consistency of sources, requiring manual escalation for conflicting data.
Capability inventory: Capabilities are restricted to read-only access for searching project sources and write access for the specific AI_Incident_Evidence_Pack_Spec.md file. It does not have access to subprocesses, network tools, or administrative commands.
Sanitization: The skill explicitly defines a redaction policy in SKILL.md and the provided template to ensure PII and tenant-sensitive data are filtered out before being included in the final handover document.

17-ai-incident-evidence-pack-spec