The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The skill uses git and curl to fetch external content from repositories and URLs defined in local configuration files. While intended for documentation, this creates a vector for ingesting untrusted data.
[COMMAND_EXECUTION] (LOW): Executes shell commands including ls, jq, test, git, and curl to manage files and download data. These operations are limited to the skill's defined workspace.
[INDIRECT_PROMPT_INJECTION] (LOW): (Category 8)
Ingestion points: External documentation files (Markdown) fetched via Git/Curl and local JSON configuration files in vault/configs/.
Boundary markers: Absent. The skill does not implement delimiters or instructions to ignore embedded prompts within the source documentation during the 'AI filtering' phase.
Capability inventory: The skill can perform network downloads, filesystem writes, and install/link new agent skills, providing a path for poisoned documentation to influence agent behavior.
Sanitization: Absent. There is no evidence of content sanitization before documentation is parsed into the generated SKILL.md.
[DYNAMIC_EXECUTION] (LOW): The skill dynamically generates and installs new SKILL.md files at runtime. This 'Skill Generator' pattern is the primary purpose of the tool but remains a risk if the source content is malicious.
[PERSISTENCE_MECHANISMS] (LOW): Uses olore install to copy generated skills to ~/.olore, maintaining these new capabilities across agent sessions.

build-docs