dsh-token-vault

The moment an AI agent needs to push to GitHub, publish to npm, or call an authenticated API, its credentials risk traversing the model context, landing in a plaintext file, or round-tripping through a browser. dsh-token-vault eliminates that exposure entirely: every GitHub, npm, or custom API token lives exclusively in the DeepSeek Harness host-level credential store. No stray files are created, no secret touches the model context, and no browser round-trip is required. The security boundary is enforced by architecture, not by convention.

The plugin registers seven agent-facing tools that cover the full credential lifecycle. Store with vault_set, migrate in one click with vault_import (from gh auth token or .npmrc), check presence with vault_has, list names with vault_list, delete with vault_remove, and—most critically—vault_run, which injects the token into a child-process environment variable (GH_TOKEN or NPM_TOKEN), executes the command, and discards the secret before stdout, stderr, or logs are ever written. The single disclosure path, vault_show, demands an explicit confirm flag, making accidental exposure by the agent structurally impossible.

If you rely on DeepSeek Harness to let an AI agent perform authenticated GitHub or npm tasks and refuse to hand tokens into a model context or scatter them across config files, this plugin was built for exactly that workflow: set credentials once in the settings panel, and every subsequent operation is brokered by the host with a clean, auditable boundary and zero plaintext residue.