PNT next-steps items 1, 4, 3: Evaluate front-door, typed report, egress lint

[richbodo]

Implements three items from plans/pnt-next-steps-plan.md (execution order 1 → 4 → 3). Items 5/6 (community-care use case, Tonsky file-sync) and item 2 (skill split) are deferred per the plan.

Item 1 — Install signpost + promote Evaluate

• README.md: the three modes now lead with Evaluate ("audit any contact app for safety before you install it"), with a concrete symlink install snippet pointing to the User's Guide.
• llms.txt: opens with a cold-agent "start with the skill" route to SKILL.md as the entry point.

Item 4 — Typed evaluate-report artifact

• tools/evaluate-report.schema.json (JSON Schema Draft 2020-12): AC-keyed findings with per-AC status (conformant/non-conformant/not-applicable/unable-to-determine), code-location citations, a summary posture, and a source-tagged (deterministic/llm/human) evidence array. Conditional rules enforce citations on (non)conformant and rationale on n/a + undetermined.
• Placed in tools/, not contracts/, because it realizes no AC (would fail lint-spec-ids.py).
• SKILL.md evaluate flow emits the artifact as source of truth; the prose report is a view over it. Makes two runs on the same candidate diffable.

Item 3 — Egress lint (deterministic AC-1 sovereignty check)

• tools/egress-lint.py: static scan for off-device egress vectors (fetch/XHR/sendBeacon/WebSocket/EventSource/import()/importScripts/axios/jQuery and HTML src/action/object data/<link href>/<use href>), flagging remote origins not on a per-flavor egress-allow.json. Ignores localhost, root-relative paths, data:/mailto:, xmlns, and <a href> navigation.
• --json emits a source: deterministic, tool: egress-lint evidence object that validates against #/$defs/evidence in the item-4 schema and folds into an AC-1 finding (verified).
• Self-test fixtures in tools/egress-lint-fixtures/{clean,dirty}, enforced by a new egress-lint-selftest CI job.

Docs

• docs/users-guide.md: signals Evaluate as the low-friction front door, makes the typed-report diff/drift workflow actionable, threads egress-lint through the build (Goal 1) and audit (Goal 2) flows, expands tool discovery, and splits the status note into tested (deterministic tooling) vs not-yet-exercised-end-to-end (build/audit/contribute skill flows).
• plans/pnt-next-steps-plan.md: items 1, 4, 3 marked done.

Verification (local)

• lint-spec-ids.py → exit 0 (12/12 contracts).
• egress-lint clean fixture → exit 0; dirty fixture → exit 1 (8 vectors); no false positives on xmlns/<a href>/mailto:/data:/localhost/root-relative.
• evaluate-report schema valid (Draft 2020-12); sample instance + all conditional negative tests pass; egress-lint --json evidence validates against the schema and drops into a full report.

🤖 Generated with Claude Code