The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied categories or questions by interpolating them directly into templates used for subagent prompts without proper boundary markers or sanitization. This allows a user to potentially override the subagent's intended research behavior.
Ingestion points: The user-supplied <app category or question> from the argument-hint in SKILL.md is used to fill the [one-line idea] and [the question] fields in the subagent brief.
Boundary markers: No delimiters or instructions are used to separate the user-provided data from the system-level instructions in the subagent brief.
Capability inventory: The skill possesses capabilities for web searching, writing to the local file system (docs/research/), and calling external tools like clone and app-brief.
Sanitization: There is no evidence of input validation or escaping for the data entering the subagent prompt.
[DYNAMIC_EXECUTION]: The instructions direct the agent to 'Turn each [U] (unverified finding) that a decision depends on into a test before you trust it (a probe script, a self-test step in tools/selftest.ps1)'. This encourages the agent to generate and potentially execute local scripts based on unverified information gathered from the web during the research phase.
[COMMAND_EXECUTION]: The skill facilitates the execution of external tools, such as clone for opening reference applications, and references the modification and use of PowerShell scripts (tools/selftest.ps1).

research