v0.5.0
trustless v0.5.0
Credential Broker CLI for AI agents — zero-knowledge secret injection. The agent never holds plaintext credentials: they are resolved at the transport (proxy) or process (subprocess) layer.
Features
- Zero-knowledge injection proxy — HTTPS_PROXY bootstrap; agents reference secrets by name only (pass / bitwarden backends)
- Subprocess run injection —
trustless runwith output sanitization, arg scanning, and policy - Integrated DLP — two-layer redaction: known-value scan (zero false positives) + 40 gitleaks-compatible pattern rules (keyword → RE2 → entropy filtering) with
pattern_mode(mask/log) and per-rulepattern_disabled - DLP hot reload — SIGHUP applies config changes (rules, mode) without restart; fail-safe on error
- Scrub commands —
scrub-db/scrub-textwith the same pattern layer (dry-run / --apply) - OAuth v1 — device flow + auto token refresh (Google, Lark), single-use refresh token handling
- Structured audit logs — JSONL events (inject/deny/spawn/redact/oauth), never contain secret values; journald or file
- Bundled agent skills — trustless-usage SKILL.md for Claude Code / Codex / Cursor / opencode / Hermes
Zero external dependencies. Single static binary. MIT licensed.
The bundled DLP rules are derived from gitleaks (MIT, Copyright (c) 2019 Zachary Rice) — see LICENSE.gitleaks / NOTICE.
Full Changelog: https://github.com/ikkun1222/trustless/commits/v0.5.0