The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax to execute a discovery script at load time: !${CLAUDE_PLUGIN_ROOT}/scripts/discover-config.sh. This executes shell commands before the agent begins processing the skill body.
[DYNAMIC_EXECUTION]: The skill dynamically loads logic from a computed path based on configuration: references/ingest-${backend}.md. This dynamic loading of instructions can be risky if the source variable is not strictly validated.
[COMMAND_EXECUTION]: The skill performs several shell-based operations including filesystem management (touch, .gitignore modification) and recursive searches through source code using grep -r to identify existing implementations.
[EXTERNAL_DOWNLOADS]: The skill initiates network activity by running git pull on configured repositories to ensure they are synchronized with the remote default branches.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external research reports and delegates their analysis to another agent, creating a surface for indirect prompt injection.
Ingestion points: The skill reads files matching patterns like *-daily-research.md, *-strategy-brief.md, and *-recommendations.md from directories specified in the configuration.
Boundary markers: The instructions include natural language directives to "preserve what the source established separately from what it recommends," but do not employ robust technical delimiters (like XML tags or specific markers) to isolate external content.
Capability inventory: The skill is granted Bash Read Write Edit Skill and performs file writes, git synchronization, and recursive searches.
Sanitization: There is no evidence of sanitization or filtering of the content extracted from external markdown reports before it is processed by the agent or delegated to the analyst harness.

ingest