The Agent Skills Directory

[CREDENTIALS_SAFE]: The skill implements robust secret management. It reads credentials only from the ATTIO_API_KEY environment variable, avoiding flags that would leak into process lists or shell history. Additionally, the scripts/attio_client.py file includes a _scrub function specifically designed to redact the API key from error messages and response bodies before they are displayed to the user.
[COMMAND_EXECUTION]: The skill uses local Python scripts to execute API calls. While it supports data-modifying commands (create, update, delete), it includes a mandatory confirmation gate (--yes flag) in scripts/cli.py. The instructions in SKILL.md explicitly direct the agent to show the proposed changes to the user and obtain verbal agreement before applying the modification flag.
[INDIRECT_PROMPT_INJECTION]: The skill has a minimal attack surface for indirect prompt injection. Although the agent processes external CRM data which could theoretically contain malicious instructions, the risk is mitigated by the enforced human-in-the-loop requirement for all write operations.
Ingestion points: CRM data is ingested via scripts/cli.py subcommands (search, query, entries).
Boundary markers: Instructions advise the agent to report raw JSON responses.
Capability inventory: Subprocess calls to the CLI for CRUD operations on the Attio API.
Sanitization: The skill relies on manual human review of output before execution of writes as the primary safety control.

attio-cli