The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions specify that the agent should load additional skills based on names found in the ## Notes section of the wayfinding map. Because the map's content can be sourced from an external issue tracker, this mechanism allows untrusted data to dynamically control which skills the agent activates at runtime.- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and acts upon data from external issue trackers (or local ticket files) which could contain malicious instructions.
Ingestion points: The agent reads issue titles, bodies, and comments to understand the destination, notes, and specific ticket questions.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its internal logic and potentially adversarial content within the tickets.
Capability inventory: The skill has the ability to read and write files, interact with network APIs (via tracker adapters), execute git commands, and load other sub-skills.
Sanitization: The skill lacks any described mechanisms for sanitizing or validating the content retrieved from the issue tracker before it is processed by the LLM.- [COMMAND_EXECUTION]: The skill requires the agent to perform git operations, specifically creating new branches and pushing them to a remote repository to persist research findings.

atlas-wayfinder