Hasp
hasp is a local secret broker designed for developers using AI coding agents. It holds credentials in an encrypted local vault and injects them into child processes at execution time, within a defined project boundary and a configurable time window. The agent never reads the raw secret value -- it receives only the resolved output of the command it is permitted to run.
AI coding agents like Claude Code, Cursor, and Aider operate with broad file system read access. Without a secret broker, API keys, database credentials, and tokens stored in .env files or environment variables are visible to the agent. hasp's spec sheet cites confirmed incidents this pattern enables: an agent deleting a production database after finding credentials in the environment, secrets leaking into published npm packages, and 28.6 million credentials exposed in git commits in 2025.
Secrets are stored in an encrypted vault using Argon2id key derivation and AEAD encryption at rest. When an application or agent requests a secret, hasp injects it at process execution time rather than exposing it as a readable value. Access grants have a ceiling of 24 hours. An append-only audit log uses tamper-evident chained HMACs, providing a verifiable record of what was accessed and when. Real-time redaction covers 11 encoding formats, preventing secrets from appearing in agent output or logs.
Local-first, no cloud dependency. Open source with source-available critical path. Signed releases include an SBOM and SLSA provenance for supply chain verification.
Placed at Assess. hasp addresses a confirmed and growing risk surface in AI-assisted development: credential exposure through agent file access. The technical approach is credible -- Argon2id, AEAD, SLSA provenance, and tamper-evident audit logs are appropriate choices. No first-person production use yet. Trial requires integrating hasp into a Claude Code workflow and verifying that secret injection works correctly without disrupting normal agent operations.
