The Agent Skills Directory

External Data Sourcing: The skill retrieves repository indices and review records from a GitHub repository (egamma/vscodesweeper-state). This external data is used to provide the agent with a snapshot of issue statuses and fix briefs. While this is the intended data source, users should ensure the external repository is appropriate for their specific environment.
Local Command Execution: The skill uses shell scripts to run system tools like gh api and jq. These commands are used to validate caches, download files, and filter JSON data, which is an expected pattern for a tool designed for data discovery and querying.
Indirect Prompt Injection Surface: The skill processes external Markdown and JSON data and presents summaries or records to the user. There is a potential for this content to contain instructions that could influence the agent's behavior, although the skill's focus on extracting specific fields and structured data reduces this surface. This finding is based on the following evidence:
Ingestion points: Data enters the context via gh api calls to the vscodesweeper-state repository (SKILL.md, sections 1, 2, and 5).
Boundary markers: The instructions lack explicit delimiters or warnings for the agent to ignore embedded instructions in the processed data.
Capability inventory: The skill uses subprocess calls to execute shell commands, GitHub CLI operations, and jq queries (SKILL.md, sections 1, 2, 5, and 6).
Sanitization: Content is partially sanitized through field extraction via jq and specific section presentation rules for review records.
Dynamic Query Assembly: The agent is instructed to dynamically generate jq filters based on natural language questions. While this enables flexible data analysis, it involves the runtime generation of processing logic applied to the ingested data.

sweeper-query