Fleetwatch 0.1.0

Updating

• One-line install (Mac mini, Raspberry Pi): run the install line again. It updates ~/fleetwatch, keeps
  your .env, sign-in and history, and restarts the service.

  curl -fsSL https://raw.githubusercontent.com/ScientiaCapital/fleetwatch/main/install.sh | bash

• Docker: docker compose pull && docker compose up -d.

• Check it: fleetwatch doctor shows the new version on its first line.

Verify the image came from this repository's CI:

gh attestation verify oci://ghcr.io/scientiacapital/fleetwatch:<version> --owner ScientiaCapital

What's Changed

Security

• security(install): pin the uv installer and default to the newest release tag by @ScientiaCapital in #70
• security(docker): pin the uv base image by digest by @ScientiaCapital in #71
• security(auth): refresh the token one process at a time by @ScientiaCapital in #72
• security(doctor): say who can read the Keychain token, and give Fleetwatch its own macOS user by @ScientiaCapital in #75
• security(auth): revoke the token on logout when Epiphan offers it (#60) by @ScientiaCapital in #77
• feat(state): store proposals and single-use approvals bound to exact arguments by @ScientiaCapital in #89
• feat(policy): describe proposable changes and accept autonomy: propose by @ScientiaCapital in #90
• feat(policy): real argument schemas for the six proposable tools by @ScientiaCapital in #92
• test(redaction): share the Anthropic key case with the Kit by @ScientiaCapital in #91
• feat(epiphan): a separate write executor that runs one approved change, once by @ScientiaCapital in #93
• feat(assistant): Claude answers fleet questions and can only propose changes by @ScientiaCapital in #94
• feat(approve): a local page where a person approves one change at a time by @ScientiaCapital in #95
• security(executor): fail closed without a write fence and allowlist sandbox devices by @ScientiaCapital in #102
• fix(state): hash-chain the audit log, bind the reason, read strictly, close the raw session by @ScientiaCapital in #103
• security(executor): validate the allowlist and harden the sandbox sign-in check by @ScientiaCapital in #106
• fix(approve): harden the approval page lockout, secret delivery and queue limits by @ScientiaCapital in #109
• fix(approve): show hidden characters in the reason and survive a lone surrogate by @ScientiaCapital in #108
• fix(approve): name --no-ai in the chat note and say when no schedule came back by @ScientiaCapital in #110

New

• feat(replay): capture live tool results as a replay sample (--capture DIR) by @ScientiaCapital in #67
• feat(replay): add a calm fixture for a screen in a quiet room by @ScientiaCapital in #69
• feat(approve): chat box on the approval page asks the assistant by @ScientiaCapital in #97
• feat(approve): show names for stream, event and channel targets by @ScientiaCapital in #104
• feat(replay): anonymize a capture into a fixture set that is safe to share by @ScientiaCapital in #113

Fixes

• fix(install): update from one release tag to a newer one by @ScientiaCapital in #79
• fix(digest): say what an offline EC20 camera really affects by @ScientiaCapital in #80
• fix(readiness): post again when a room's verdict changes before the event by @ScientiaCapital in #81
• fix(policy): per-action disruptive rules and per-tool lead time by @ScientiaCapital in #105
• fix(epiphan): read a result that doesn't match Epiphan's own output schema by @ScientiaCapital in #111
• fix(parse): read the status list a live system-status call returns by @ScientiaCapital in #112
• fix(notify): label console output as a preview, not a post by @ScientiaCapital in #117
• fix(deploy): flush printed digests under launchd and systemd by @ScientiaCapital in #116
• fix(ci): attest the image under its lower-case GHCR name by @ScientiaCapital in #118

Docs

• docs(agents): keep internal plans out of the public repo by @ScientiaCapital in #66
• docs(readme): Epiphan writing style, language switcher, calm demo and --capture by @ScientiaCapital in #73
• docs: follow the Epiphan writing style in SECURITY, CONTRIBUTING and the docs site by @ScientiaCapital in #74
• docs(readme): add an Updating section in English and Spanish (#14) by @ScientiaCapital in #76
• docs: make README, CLAUDE.md and AGENTS.md match what main does today by @ScientiaCapital in #83
• docs: say exactly what read-only and redaction cover by @ScientiaCapital in #87
• docs(design): v0.2 assistant that proposes changes a person approves by @ScientiaCapital in #86
• docs(v0.2): trust model and approved-writes flow by @ScientiaCapital in #96
• docs: bring CLAUDE.md and AGENTS.md up to what main does today by @ScientiaCapital in #115
• docs(approve): describe the lockout, the secret file, the queue and deny all by @ScientiaCapital in #114
• docs: the soak is running; list the Sprint 3 review findings by @ScientiaCapital in #125

Everything else

• chore(auth): remove the legacy token file once the new store holds a token by @ScientiaCapital in #68
• ci: fix pipefail, arm64 builds, latest tag, and the Python badge by @ScientiaCapital in #84
• ci: mark hyphenated tags as pre-releases on GitHub by @ScientiaCapital in #88
• test(assistant): injection corpus for hostile fleet names by @ScientiaCapital in #107
• fix(redact): mask dotted, spaced and paired secret names and webhook URLs by @ScientiaCapital in #128
• fix(run): count a stray cancel from the MCP client as a failed beat by @ScientiaCapital in #130
• docs: the soak is done; the agent is stopped and signed out by @ScientiaCapital in #131
• feat(cli): add fleetwatch connect, a guided first run by @ScientiaCapital in #132
• docs: the Now section names fleetwatch connect by @ScientiaCapital in #133
• docs: fix stale status, sign-in and link text; point installer at connect by @ScientiaCapital in #136
• docs: add a command reference and document the v0.2 propose rules by @ScientiaCapital in #137
• test: cover room state, the approval read, readiness rules and the closed-client guard by @ScientiaCapital in #138
• docs: document the release steps and supported versions; test the version matches by @ScientiaCapital in #141
• refactor: one Priority.rank, redacted failed-read log, reasons on bare noqa by @ScientiaCapital in #139
• refactor(scanner): split scan() into one helper per rule by @ScientiaCapital in #140
• feat(policy): put the group tools on the read and write lists; design the group fence by @ScientiaCapital in #142
• fix(readiness): say a room is already recording before its event starts by @ScientiaCapital in #143
• docs: the Now section names the group tools and their design by @ScientiaCapital in #144

Full Changelog: https://github.com/ScientiaCapital/fleetwatch/commits/v0.1.0