The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data, including issue descriptions, git commit history, and source code content, which presents a potential surface for indirect prompt injection.
Ingestion points: The agent ingests user-provided <issue> text, $ARGUMENTS, and results from git log, git blame, and Grep search operations (SKILL.md).
Boundary markers: The skill includes a TERMINATION_CONTRACT intended to restrict the capabilities of sub-agents spawned during the archaeology and generalization phases (SKILL.md).
Capability inventory: The skill has access to powerful tools such as Bash, Write, Edit, and spawn_agents_on_csv, allowing it to modify files and execute shell commands (SKILL.md).
Sanitization: No explicit content sanitization or escaping mechanisms for untrusted data were observed in the provided instructions.
[COMMAND_EXECUTION]: The skill uses Bash to perform standard development tasks, including repository history analysis and file searching. It also uses spawn_agents_on_csv to delegate analysis tasks to sub-agents using predefined templates and instructions, which are constrained by a termination contract to prevent unauthorized tool usage by sub-agents.

odyssey-debug