The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes HAR (HTTP Archive) files, which are external data sources that could theoretically contain malicious content designed to influence agent behavior.
Ingestion points: Instructions in SKILL.md and references/har-extraction.md describe loading and parsing JSON-based HAR files from temporary directories.
Boundary markers: The skill incorporates strict logical guardrails and non-negotiable authorization rules in SKILL.md, though it does not define specific character-based delimiters for the data input.
Capability inventory: The skill leverages curl and standard HTTP client libraries (e.g., Python requests) to execute reconstructed API calls across various scripts.
Sanitization: The workflow enforces a selective reconstruction process where only specific, necessary fields (method, URL, minimal headers) are extracted, effectively sanitizing the input by excluding extraneous or potentially malicious captured data.
[SAFE]: The skill promotes defensive security practices, such as applying restrictive file permissions (chmod 600) to sensitive artifacts, avoiding the hardcoding of session tokens, and mandating independent verification of replayed requests to ensure they do not rely on fragile or compromised session state.

har-api-reverse-engineering