The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands such as git blame, git log, and gh pr view to establish a 'Code Anchor' for its investigation. These tools are used to extract commit history, diffs, and pull request metadata, which are necessary for the skill's archaeological function.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various external platforms, creating a potential vector for indirect prompt injection where malicious instructions stored in tickets, chat messages, or documentation could influence the agent.
Ingestion points: Data is ingested via Model Context Protocol (MCP) tools from Slack, Linear, Notion, Datadog, Sentry, and Databricks as described in SKILL.md and the category playbooks in references/sources/.
Boundary markers: Explicit structural delimiters are absent, but the references/epistemics.md and references/investigator-prompt.md files mandate a strict posture of quoting evidence verbatim and using confidence-weighted tiers to separate fact from inference.
Capability inventory: The skill has the capability to execute shell commands (git, gh) and perform extensive read-only queries against integrated enterprise databases and APIs as detailed in SKILL.md.
Sanitization: There is no programmatic sanitization of ingested text; the skill relies on a 'Synthesizer' subagent to logically evaluate and cross-reference findings before presentation to the user.

why