Skip to content

docs: fix the fact-check items of #163 - #164

Merged
DavidGodefroid merged 1 commit into
mainfrom
docs/fact-check-163
Nov 15, 2026
Merged

DavidGodefroid merged 1 commit into
mainfrom
docs/fact-check-163

Conversation

@DavidGodefroid

Copy link
Copy Markdown
Owner

Fixes the docs items of the fact-check in #163: README, SECURITY.md, docs/**, the site (site/**, demo included) and the texts shipped to agents (internal/agentassets/agents.md, skill.md, the locklog_scan tool description) now say what origin/main (v0.4.0) does. Docs only; no behaviour change.

Each item was re-verified against the code before the fix. Highlights:

  • v0.4.0 is the confined-agents release (Bump lockkit to v0.20.0: refuse unconfined agents by default (lockkit#133) #158). The ES/OS incident gates and a masked export are "planned", with no version (items 1, 2).
  • lox add vs doctor --fix (items 18, 20, 39), settled against lockhub origin/main: the guided lox add locklog runs locklog profile add, locklog init claude and locklog doctor --json, without --fix (internal/loxinit/doctorfix.go, DoctorArgs: locklog is not in doctorFixes; cmd/lox/add_test.go TestAddLocklogNonInteractive). lox add locklog <name> (TYPE INSTANCE) only declares the lock. The docs now use the guided form and say both. The locklog formula alone needs lox (item 6).
  • Masking and audit claims say "what locklog masks", with links to "what is not masked" and retention (items 3, 30). Invariant 11 is a cap (E_BUSY), not a queue (31). The Landlock write and read lists are complete (32). lockkit's untrusted-data header is quoted the same way everywhere (41).
  • Agent assets: approval is a ticket on lockhub, not at the console; no "hub mode" (item 9).
  • HolmesGPT recipe: tickets can be auto-approved when a profile's auto rules say auto (item 4).
  • performance.md, loki.md, sources.md: scan pacing, metric-query cap, calibration corpus, index rate, the perf job, the JSON profile and the spike to-dos are restated against the code (items 7, 8, 10-14, 22, 25-29).
  • Site: no hard-coded version in footers or badges (5), invariant 17 and a "Confined agents" section on the security page (16), the FAQ unmask wording (17), a softened comparison column (40), base64/hex in the limits (38). The demo's denial text and the unmask exposure line now follow the code; demo.svg was regenerated (36, 37).
  • Item 42 (locklog.dev URLs) needed no change.

Checked locally: go test ./internal/agentassets/... ./internal/mcptools/... ./cmd/locklog/..., scripts/check-site.py site (6 pages, 0 errors), scripts/brand/check_demo.py, scripts/hygiene.sh.

Closes #163

🤖 Generated with Claude Code

Make README, SECURITY.md, docs/, the site and the texts shipped to agents
say what origin/main (v0.4.0) and lockhub's lox do:

- v0.4.0 is the confined-agents release; the ES/OS incident gates and a
  masked export are planned, with no version.
- lox add locklog (guided) runs profile add, init claude and doctor
  without --fix (lockhub internal/loxinit DoctorArgs); lox add locklog
  <name> only declares the lock. The locklog formula alone needs lox.
- Masking and audit claims say "what locklog masks", with links to what
  is not masked and retention; invariant 11 is a cap (E_BUSY), not a
  queue; the Landlock write list is complete; lockkit's untrusted-data
  header is quoted the same way everywhere.
- Agent assets and the locklog_scan description: approval is a ticket on
  lockhub, not at the console; no "hub mode" (the only mode).
- performance.md, loki.md, sources.md: stale budgets, calibration, index
  rate, perf job and spike to-dos restated against the code.
- Site: no hard-coded version in footers and badges, invariant 17 and a
  confined-agents section on the security page, FAQ unmask wording,
  softened comparison column; the demo's denial text and exposure line
  follow the code (demo.svg regenerated).

Closes #163

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@DavidGodefroid
DavidGodefroid merged commit ae7ca09 into main Nov 15, 2026
26 checks passed
@DavidGodefroid
DavidGodefroid deleted the docs/fact-check-163 branch November 15, 2026 13:05
@DavidGodefroid DavidGodefroid mentioned this pull request Nov 15, 2026
DavidGodefroid added a commit that referenced this pull request Nov 15, 2026
Release v0.4.1: the texts shipped to agents follow the code (#163,
#164).

## What changes

- `CHANGELOG.md`: `[0.4.1] - 2026-11-15` section and compare links.

There is no other version label to bump: #164 removed the hard-coded
versions from SECURITY.md, docs/install.md and the site.

Why a release: `internal/agentassets/agents.md`, `skill.md` and the
`locklog_scan` tool description are embedded in the binary and the
image, so agents only get #164's fixes with a new release.

After merge: tag `v0.4.1` on main. `release.yml` drafts the release with
GoReleaser, signs it, publishes it, then dispatches pages, the Homebrew
tap and the lockfamily docs.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Docs fact-check: wrong or misleading claims

1 participant