The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a provided Python script (scripts/validate-near-miss.py) to validate the structure of near-miss reports. This script is a local utility that performs basic text matching to ensure required sections are present before the report is processed further.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and analyze agent failures. This creates a surface for indirect prompt injection because the incident reports being reviewed may contain the original malicious payloads or instructions that triggered the failure.
Ingestion points: The skill reads and processes near-miss reports (near-miss.md) created during the review process.
Boundary markers: The skill uses structured markdown headers from a template (e.g., '## 観測事実', '## 再現条件') to organize data, but it lacks explicit 'ignore embedded instructions' delimiters within those sections.
Capability inventory: The skill possesses the ability to read and write local files and execute a Python script to validate report content.
Sanitization: The validation script performs basic checks for missing headers but does not sanitize or filter the content within the report sections for potential instructions or malicious scripts.

agent-near-miss-review