The estate answers, the page does not: status.sh, preflight.sh, requires:, reasoning traps

[Eslam93]

What this adds

The kit asked "what is deployed, when the project rule names a way to check" and left the answer as prose. This makes the estate answer, and gives runbooks a way to prove their preconditions.

status.sh runs two project-owned inputs: .claude/tools/status.project.sh, the project's read-only probes (one finding per line: level, section, subject, detail, remedy), and .claude/deadlines.conf, only the dates no system can answer, each with a lead time and an owner. RED first, exit 1 on any RED, a line it cannot parse is a RED of its own, RED and AMBER must carry a remedy. /orient runs it before anything else and leads with the RED lines. --canary must fail.

preflight.sh proves, read-only, the capabilities a page names in its requires: header field, from 00-orientation/capabilities.md: one row per token with a blast marker, what it grants, a proof command, the expected output, how to obtain it. UNRESOLVED is reported, never invented around. knowledge-check.sh refuses a token the page does not define. /work proves a page's tokens before the one yes; a 🔴 token makes the task Tier 3.

working-here.md gains a Reasoning section: the size of a mechanism reported as its effect, a green test on an unreachable path, the dangerous zero, an instrument that measures itself, log volume read as usage.

The failures that paid for it (one installation, dates in the investigation page)

• 2026-08-24: a spot-check of seven written operational claims found five stale or wrong, every one computable from a live system.
• 2026-08-01: a required-reviewer gate recorded a non-member's vote and stayed blocked, so tooling that votes reported success while nothing moved.
• 2026-08-24: a fix recorded as proven by replay (12 of 77 to 44 of 77) had fired zero times in production.

How it was shown to go red

• estate.test.sh: 32 fixture cases against the real tools with --root, 32 passed.
• knowledge-check.test.sh: three new cases for requires:, 74 passed.
• verify.sh: 24 passed; --canary fails as it must; the status canary is part of both.
• verify.sh --hooks on the owner's Windows machine (Git Bash + PowerShell 5.1), 2026-09-22: hooks.test.sh 219 passed, knowledge-check.test.sh 74 cases, estate.test.sh 32 cases, install.test.sh passed; 28 checks, 0 failed.
• CI (Linux) runs the same three jobs on this pull request.

Files

• new: .claude/tools/status.sh, .claude/tools/preflight.sh, .claude/tools/estate.test.sh, docs/knowledge-base/_investigations/2026-09-22-estate-checks/methodology.md
• changed: verify.sh, knowledge-check.sh (+ its test), standing-orders.md, working-here.md, knowledge-base.md, orient and work skills, README.md, START-HERE.md, decisions.md (D-22), index.md, 99-pending.md

Limits are stated in the investigation page: both tools are bash, like every tool here; proof commands are run as written, so a capabilities page is trusted like verify.project.sh; the rule about values in prose stays advice.

🤖 Generated with Claude Code