The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a coordination pattern where a lead agent synthesizes findings from multiple subagents (documented in modules/findings-format.md and modules/subagent-coordination.md). This creates an attack surface for indirect prompt injection via the findings files processed by the parent agent.
Ingestion points: The lead agent is instructed to read summaries from .coordination/agents/*.findings.md files.
Boundary markers: While using structured Markdown, the skill does not explicitly use markers to ignore instructions within the processed text.
Capability inventory: Includes bash hooks, Task tool for subagent spawning, and local file writing.
Sanitization: Input filtering using jq is present in hooks, but no specific sanitization is defined for the synthesis of findings content.
[COMMAND_EXECUTION]: The SKILL.md frontmatter defines several hooks (PreToolUse, PostToolUse, Stop) that execute bash commands to log context usage metrics and tool activity to a local file (/tmp/skill-audit.log).

context-optimization