Integrations¶

SkillRoute¶

SkillRoute and toolprint cover the two halves of the same problem. Both deal with text an agent reads and then obeys:

	SkillRoute	toolprint
Substrate	`SKILL.md` bundles	MCP capability definitions
Question	Which one should the agent use?	Should the agent trust it at all?
On skills	Indexes, routes, validates against the spec	Scans for injection, pins against rug-pulls
Overlap	none — spec compliance is not a security check

They integrate in both directions.

SkillRoute → toolprint: knowing where every client lives¶

toolprint's own auto-discovery covers four project-local paths. SkillRoute maintains manifests for 15 agent harnesses across three platforms, including where each keeps its MCP config.

That table is vendored into src/connect/clients.ts and powers --all-clients:

npx toolprint scan --all-clients

It is vendored, not imported, so toolprint keeps its "no install, no Python" promise and works whether or not SkillRoute is present.

If you do have the skillroute CLI, --use-skillroute layers its live detection on top:

npx toolprint scan --all-clients --use-skillroute

This runs skillroute harness detect --json, which reports which harnesses are actually installed and their resolved config paths — picking up clients added to SkillRoute after your toolprint was published. Because it executes a third-party binary, it is opt-in, the same way --probe is. A missing or unexpected response degrades to the built-in table and says so.

toolprint → SkillRoute: security for skill bundles¶

SkillRoute's validate answers "is this bundle well-formed?" toolprint answers "is it safe, and is it still the one you reviewed?" Run both:

skillroute validate ./skills --strict     # spec compliance
npx toolprint scan --skills ./skills      # injection + rug-pull

See Skill Bundles for what the security pass actually catches.

Scanning SkillRoute itself¶

SkillRoute ships an MCP server, so it is a scan target like any other:

npx toolprint scan npx:@skillroute/mcp-server
npx toolprint pin  npx:@skillroute/mcp-server

MCP clients¶

Any client whose config toolprint can read is covered — see the client list. Adding one is a data change in src/connect/clients.ts; see CONTRIBUTING.md.

GitHub¶

Code scanning — --sarif emits SARIF 2.1.0 for github/codeql-action/upload-sarif.
Pull requests — comment-on-pr: true upserts a sticky findings table.

Both in CI.

Anything else¶

--json is a stable, documented contract — see JSON Schema. Every finding carries a stable id, so dashboards and baselines can track one finding across runs.