The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts using shell commands to manage orchestration workflows.
Evidence: Phase 2 in SKILL.md executes python agent-orchestrator/scripts/scan_registry.py, match_skills.py, and orchestrate.py.
Risk: User input is interpolated directly into the command line: python agent-orchestrator/scripts/match_skills.py "<tarefa do usuário>". Without explicit sanitization by the environment, this presents a command injection surface if the user-provided task description contains shell metacharacters.
[PROMPT_INJECTION]: The skill facilitates an orchestration pattern that propagates untrusted user data to multiple specialized agents, acting as a surface for indirect prompt injection.
Ingestion points: User-provided task descriptions enter the context in Phase 2 and are passed as queries to agents in Phase 3.
Boundary markers: Absent. The protocol does not use delimiters or instructions to ignore embedded commands within the user task string.
Capability inventory: The skill triggers shell execution (python), file system operations (open), and potentially network-bound package installations (pip).
Sanitization: No sanitization or validation of the task input is specified in the instruction files.
[EXTERNAL_DOWNLOADS]: The skill documentation references the use of external package managers for dependency resolution.
Evidence: references/problem-catalog.md includes pip install -r requirements.txt as a preventative step for script execution errors, which involves fetching packages from remote registries.

task-intelligence