swamp

Model any API with Swamp, test it, and enrich *Claw with new capabilities — full lifecycle from idea to working integration

This skill largely does what it says (it documents how to use the 'swamp' CLI to model APIs and export models into new *Claw skills), but there are a few things to check before installing or running it: - Verify the 'swamp' binary: ensure it comes from a trusted source (e.g., the official GitHub repo mentioned) and is the version you expect. The skill requires that binary to exist on PATH. - Review .claude/settings.local.json: the file that ships with the skill only whitelists a few swamp subcommands but the SKILL.md expects broad use of many commands. Confirm what commands your agent will actually be allowed to run and reconcile any permission gaps. - Be cautious with authentication flows: 'swamp auth login' may open a browser or accept username/password flags. Avoid entering credentials into unfamiliar registries; prefer browser/OAuth flows and verify the SWAMP_CLUB_URL before logging in. - Audit generated artifacts: the workflow explicitly instructs exporting models and generating SKILL.md (new skills) and pushing/pulling extensions. Always review any generated SKILL.md, extension manifest, or packaged extension before installing or publishing — these artifacts can add new capabilities to your agent. - Use least privilege and sandboxing: run initial experiments in a sandboxed environment or VM, and avoid granting broad system access or persistent install rights until you have validated the tool and its outputs. If you want a stronger conclusion, provide the actual 'swamp' binary source (checksums or install instructions) and the agent's runtime permission policy so I can re-evaluate the whitelist/permission mismatch and the potential blast radius of auto-published extensions.

Type: OpenClaw Skill Name: swamp Version: 0.2.0 The skill is classified as suspicious due to the broad capabilities granted to the AI agent, specifically the ability to interact with a remote extension registry. While the `.claude/settings.local.json` file restricts the agent to only `swamp auth` and `swamp extension` commands, these still pose significant risks. `swamp auth login` allows network communication for authentication, potentially exposing credentials if misused. More critically, `swamp extension pull` enables the agent to download arbitrary content from a remote server, and `swamp extension push` allows the agent to publish content to a public registry, creating a direct supply chain attack vector if the agent is prompted to upload malicious extensions. The `SKILL.md` also describes highly privileged operations like executing arbitrary shell commands via `swamp model method run` and publishing new skills to ClawHub, which are not currently permitted but highlight the inherent power and potential for misuse of the underlying `swamp` tool.