The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill manages Odoo connection profiles and provides playbooks for reading local server configuration files (odoo.conf) which contain sensitive credentials.
Evidence: scripts/odoo_common.py manages ~/.config/odoorpc/config.yaml for storing Odoo passwords, and the skill includes instructions for analyzing local Odoo configuration files.
Mitigation: The skill automatically sets restricted filesystem permissions (0600) on its configuration files and includes a redact_profile helper to ensure passwords and API keys are not leaked in command output.
[INDIRECT_PROMPT_INJECTION]: As a tool that processes data from remote Odoo databases, the skill has an inherent surface for indirect prompt injection if records contain malicious instructions.
Ingestion points: Record data (names, descriptions, values) fetched via RPC in odoo_query.py and odoo_mutate.py.
Boundary markers: The skill does not currently define specific delimiters or instructions to ignore embedded prompts within retrieved Odoo data.
Capability inventory: Includes the ability to write files (snapshots and module descriptions), execute local scripts, and mutate remote Odoo data.
Sanitization: The skill uses structured JSON for domains and RPC calls but does not perform content-level sanitization of record data presented to the agent.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic method invocation to interact with Odoo models.
Evidence: scripts/odoo_query.py uses getattr(Model, args.method) to call model methods based on command arguments.
Mitigation: Execution is guarded by a mandatory --i-know-method-is-readonly flag to prevent the accidental call of state-changing methods through the query interface.

odoo-agent-skill