Add Agent Team Readiness scan core

[Gandy2025]

Summary

• add the agent-team-readiness skill and a deterministic, read-only repository scanner
• score six fixed dimensions under the published atr-1 evidence contract
• emit a machine-readable report, evidence inventory, review-only AGENTS.md draft, and Context Tree seed map
• preserve repo-observable unknowns instead of inferring runtime isolation, permissions, or team behavior

Safety and determinism

• never execute commands discovered in the scanned repository
• disable repository-configured Git execution surfaces such as fsmonitor
• reject output inside the target and refuse symlink/hard-link overwrites
• skip symlinked, deleted, unreadable, hollow, placeholder-only, and non-authoritative fixture evidence
• redact bounded credential-like evidence before serialization
• pin remote scans to an explicit commit and label dirty local scans without claiming HEAD

Validation

• npm test — 38/38 passing
• npm run check
• npm audit --audit-level=high — 0 vulnerabilities
• skill structure validation — passing
• 22 adversarial/positive-control fixture repositories validate against both the executable validator and JSON Schema
• 12 pinned public repositories, 72/72 per-dimension human band agreements, 0 disagreements
• real pinned First Tree scan: 84/100 headline score, 99% evidence coverage
• final independent correctness/security review: blocker-free

Artifacts

The checked First Tree example includes all four outputs under examples/first-tree/. Calibration keeps the full evidence, unknowns, human notes, and implementation/input digests so a stale baseline fails CI.

[Gandy2025]

Superseded by the broader, already-merged core implementation in #1 (44 tests / 25 fixtures). Closing this duplicate to keep main on the stronger reviewed version.