Secure Gmail

Secure Gmail skill using Composio brokered OAuth — no raw tokens stored locally. Reads, searches, and drafts emails with least-privilege enforcement. Blocks...

Do not install or run this skill until the contradictions are resolved. Specific actions to request from the author or check yourself: - Ask the author to remove GMAIL_SEND_EMAIL from the session create() enable list in agent.py if send capability is truly disallowed. The current code enables send despite documentation saying send is blocked. - Ask for a clear statement of which Composio tool slugs are actually allowed on the Composio dashboard (and confirm the dashboard policy blocks GMAIL_SEND_EMAIL). Documentation alone is not sufficient; confirm the brokered-policy configuration. - Ensure registry metadata matches SKILL.md: COMPOSIO_API_KEY should be declared as a required env var in the skill manifest/registry. - Consider avoiding storing COMPOSIO_API_KEY in a local .env; if you must, ensure file permissions are restrictive and the key has least privilege. Prefer a secrets manager if available. - Request an update to GMAIL_TOOL_SLUGS.md to remove or clearly mark high-privilege slugs that this skill will not request, so reviewers can't be misled. - For extra caution, run the skill in a restricted environment and monitor Composio logs (as SKILL.md suggests) to verify that no send/delete actions are issued. If the author cannot satisfactorily explain and fix the send-slug discrepancy and the metadata mismatch, treat the skill as untrusted.

Type: OpenClaw Skill Name: secure-gmail Version: 0.1.0 The skill is classified as suspicious due to a critical discrepancy between its advertised security posture and its actual code implementation. While the `SKILL.md` and `README.md` documentation explicitly state that the skill is 'read-only + draft' and that 'GMAIL_SEND_EMAIL' is blocked at the Composio API gateway, the `agent.py` code explicitly enables the `GMAIL_SEND_EMAIL` capability when creating the Composio session. This creates a severe vulnerability: if the Composio gateway's blocking mechanism fails or is overridden by the `enable` list, the skill could be coerced into sending emails, directly contradicting its stated purpose and security claims. This is a risky capability enabled in `agent.py` that directly undermines the security assurances provided in `SKILL.md` and `README.md`.