Failproof AI 1.0.2

Failproof AI 1.0.2

Failproof AI 1.0.2 makes policy enforcement safer, introduces policy packs as the standard distribution model, and rebuilds the documentation around anyone running agents—not only developers.

Policy packs

policies is now the main CLI noun. The package does not include an installable policy pack, so you explicitly choose which policies to run. The always-on block-failproofai-commands guard remains active to stop agents from disabling Failproof AI itself.

Policy packs are pinned to their artifact digest and source commit. If a pack is broken, missing or cannot be verified, Failproof AI denies only within that pack's declared scope instead of silently dropping enforcement.

Shared artifacts, renamed packs, policy selections and agent scopes now remain consistent across installation, loading and removal.

Safer enforcement

The always-on state guard now catches more ways an agent could modify or delete ~/.failproofai, including shell globs, interpreters, variables, loops, pipes and agent-native Write, Edit and NotebookEdit operations.

The same work removes false positives around ordinary reads and commands that inspect Failproof AI state without changing it.

Observe mode

Policies can be evaluated in observe mode before enforcement. The policy runs normally and its verdict is recorded, but the agent is not blocked. This works for locally published packs and Cloud deployments, and the results can be inspected through policy activity or FailproofAI Cloud.

Publishing and management

failproofai publish now verifies every bundled source against the commit named by the release, including imported helpers, ignored files, renamed files and symlinked paths.

Publishing to an existing private repository is refused by default because anonymously installed packs cannot fetch private release assets. Pack installation and removal also handle ambiguous names, version suffixes, case differences and identical artifacts more safely.

Audits and dashboard

failproofai audit can inspect supported agent history locally and open findings at http://localhost:8020/audit. Analysis and findings stay on the machine; telemetry and scheduled-report behaviour are now documented separately and accurately.

Audit resume handling no longer double-counts findings when a transcript has been rewritten or truncated. The dashboard also correctly scopes policy toggles and parameters to the selected pack.

Documentation overhaul

The README and all 68 English documentation pages were audited against the shipped 1.0.2 CLI.

The rebuilt docs now cover local and Cloud setup, policy packs, observe mode, local audits, session tracing, policy authoring and fleet deployment. Enforcement differences across all twelve supported agent harnesses are documented instead of assuming every harness behaves identically.

Upgrading

npm install -g failproofai@1.0.2
failproofai config
failproofai policies add FailproofAI/policies

For non-interactive Cloud setup, use failproofai config --token <machine-key> instead.

Full Changelog: v1.0.1...v1.0.2