The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted code diffs to identify new functions, constants, and types. This data is then used to influence the agent's search behavior and repository interactions.
Ingestion points: The skill explicitly processes "any diff that adds a function, constant, Annotated type, validator, enum, model property, repository method, React component, or hook" as specified in the Scope section of SKILL.md.
Boundary markers: There are no explicit instructions or delimiters defined to separate the untrusted diff content from the agent's search logic or to instruct the agent to ignore instructions embedded within the diff.
Capability inventory: The agent uses rg (ripgrep) via subprocess calls to search the server/polar and clients/packages directories.
Sanitization: The instructions do not include requirements for escaping or validating strings extracted from the diff before they are passed to the shell-based search tool.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands using ripgrep based on strings found in the input diff. The instruction rg -n "def <name>|<Name>\b" server/polar clients/packages creates a path for command injection if a malicious diff contains symbol names with shell metacharacters (e.g., backticks, semicolons, or command substitutions).

reuse-check