The Agent Skills Directory

[PROMPT_INJECTION]: The skill collection includes a specialized security auditing component (ai-security) which contains numerous prompt injection and jailbreak payloads in its metadata and reference files (SKILL.md, references/atlas-coverage.md). Examples include 'DAN mode' framing, system prompt extraction requests (e.g., 'Repeat your initial instructions verbatim'), and behavior override markers. While these are intended for test coverage and pattern matching within a security scanner, they represent literal injection strings that an agent may mistakenly interpret as instructions.
[DYNAMIC_EXECUTION]: Multiple scripts within the collection implement dynamic code or command execution. The google-workspace-cli/scripts/gws_recipe_runner.py script executes shell commands using subprocess.run(cmd, shell=True). The commands in the recipe dictionary contain placeholders (e.g., {to}, {subject}) intended for substitution. Additionally, code-reviewer/scripts/review_report_generator.py uses sys.executable to dynamically invoke other Python scripts (pr_analyzer.py and code_quality_checker.py) located in the same directory.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process a wide variety of untrusted external content, creating a significant surface for indirect prompt injection attacks.
Ingestion points: Data enters the agent's context through several scripts, including a11y_scanner.py (scanning project source files), pr_analyzer.py (analyzing git diffs), timeline_reconstructor.py (processing external JSON event logs), and ai_threat_scanner.py (reading user-provided test files).
Boundary markers: The analyzed scripts do not use specific delimiters or instructions to treat the ingested data as untrusted or to ignore embedded natural language commands.
Capability inventory: The skill possesses powerful system capabilities, including the ability to execute shell commands (via git, gws, or powershell) and recursive filesystem access.
Sanitization: Ingested content is processed as raw text without sanitization or filtering to remove potential instructions before presentation to the agent.
[COMMAND_EXECUTION]: The collection provides tools that generate and execute complex CLI commands and system administration scripts. The google-workspace-cli skill executes gws commands to manage Gmail, Drive, and Admin settings, while the ms365-tenant-manager skill generates PowerShell scripts for high-privilege Microsoft 365 Global Admin tasks. These capabilities provide a direct path for an agent to perform invasive system operations if manipulated.
[EXTERNAL_DOWNLOADS]: The documentation in README.md and SKILL.md recommends installing components via npx ai-agent-skills install afaizalam2003/... from a custom repository. The skill also generates HTML templates that import external JavaScript libraries like GSAP from the jsDelivr CDN (https://cdn.jsdelivr.net/npm/gsap@3.12.5/dist/gsap.min.js).

engineering-skills