The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from various third-party platforms including Git, Linear, Notion, Slack, Datadog, Sentry, and Databricks.
Ingestion points: Data is pulled from external repositories, ticket descriptions, chat messages, and database records via MCP tools as defined in the references/sources/ playbooks.
Boundary markers: There are no explicit technical boundary markers or sanitization routines for the data retrieved from these sources.
Capability inventory: The skill utilizes shell commands (git, rg), SQL execution (execute_sql_read_only), and various specialized search tools across the integrated platforms.
Sanitization: The skill relies on instructional guardrails, such as the references/epistemics.md framework and the requirement to 'Spot-check citations' in the references/synthesizer-prompt.md, to mitigate the influence of biased or malicious data.
[COMMAND_EXECUTION]: The skill provides the agent with templates for shell-based investigation using tools like git and ripgrep (rg), as well as SQL-based analysis. While these are core functionalities for the skill's purpose, they rely on the agent to correctly escape and handle parameters derived from user input or external data, presenting a potential injection surface.

pstack-why