The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local shell commands to establish historical context for its investigation.
Evidence: SKILL.md (Step 2) and references/sources/code-archaeology.md specify the use of git blame, git log, and the GitHub CLI (gh pr view) to extract commit history, PR bodies, and discussion threads.
[INDIRECT_PROMPT_INJECTION]: The skill ingests large volumes of untrusted, externally-sourced data which could contain malicious instructions designed to subvert the agent's logic during synthesis.
Ingestion points: External data is fetched from diverse sources including GitHub PR comments, Linear issue descriptions, Notion documentation, Slack chat threads, and Datadog/Sentry logs.
Boundary markers: The investigator-prompt.md and synthesizer-prompt.md files establish clear roles and instruct investigators to "Gather, don't conclude" and "Quote verbatim," providing some logical separation between source data and agent instructions.
Capability inventory: The skill has significant capabilities, including the ability to spawn child agents (delegate_task), execute git/gh commands, and interact with numerous third-party MCP tools for database and API access.
Sanitization: There is no explicit sanitization, filtering, or escaping of retrieved external text (such as PR comments or chat messages) before it is processed by the synthesizer sub-agent.

why