The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process raw stack traces and error logs, which are untrusted data sources often generated by third-party code or external environments.
Ingestion points: The condense_traceback Python function in SKILL.md accepts a raw_trace string, and the agent is instructed to process all test/build failures.
Boundary markers: The protocol uses a standardized signature 🚨 [FAIL: <file>:<line>], but there are no explicit instructions for the agent to treat the summarized content as potentially untrusted or to ignore instructions embedded within the error message.
Capability inventory: The skill is restricted to text processing and does not request tools for network access, file writing, or subprocess execution beyond the summarization logic.
Sanitization: The extraction logic relies on regex to identify file paths and error types, providing some structural filtering, but it does not sanitize or escape the direct failure message content before interpolation.
[DYNAMIC_EXECUTION]: The skill includes a Python implementation for a condense_traceback function.
The code uses standard library regular expressions to parse stack traces and does not utilize dangerous functions like eval(), exec(), or runtime compilation of external code.
The logic is a static template for processing local text data, posing minimal risk in its current form.

single-line-error-summaries