DevThrottle

Evidence shows per-user install roots (%LOCALAPPDATA%\cc-director, ~/Applications), explicit no-admin claims, and SHA-256 verification before placement, so least_privilege scores 2. user_confirmation scores only 1: the MISSION doc admits the macOS installer runs rm -rf over a running Director's bundle with no gate (item 3.1 explicitly deferred), and updates are described as silent with no prompt or UAC. data_flow_transparency scores 1: README admits inference always routes through DevThrottle with no BYOK, but no file describes what is transmitted or retained. sensitive_data_handling scores 1: SECURITY.md concedes credentials sit in the user profile and defers protection to the OS, with no encryption or isolation evidence. dependency_security scores 1: package.json pins only eslint/typescript-eslint, with no audit or lockfile evidence. external_effects scores 1: the installer writes PATH, registers autostart, and launches resident processes, and the HANDOVER documents orphan processes and overwriting a running app without adequate guards. rollback scores 2: INSTALLATION.md describes .old backups, a rollback command, and update-pins, but explicitly states there is no automatic health check or auto-rollback. source_attribution scores 1: publisher is unverified and the code-signing subject (Center Consulting Inc.) is not tied to the repository owner in any file.

self_consistency scores 1: documents contradict each other — INSTALLER_HANDOVER says the Director is framework-dependent and needs .NET 10 while INSTALLATION.md says it is now self-contained; HANDOVER says uninstall deletes all logs while MISSION corrects this to an opt-in data wipe; version numbers jump between v0.6.3, v1.8.4, v1.8.5 and v2.28.0. dependency_availability scores 2: INSTALLATION.md documents an offline --release-dir path, self-contained binaries, and a SHA manifest, though none is verified here. failure_messages scores 1: MISSION 2.1 admits the failure reason is computed and discarded so the UI shows only 'Failed', marked DONE without code evidence, and HANDOVER records a silently swallowed registration failure.

audience_and_scenarios scores 2: README and install-prompt clearly separate human users, coding agents, and Windows/macOS scenarios. capability_boundaries scores 2: macOS is stated as Director-only and unable to host a self-hosted gateway, and the gateway is a Windows tray app, so boundaries are stated. trigger_precision scores 1: install-prompt relies on stable.json plus SHA checks, but the CI config is truncated and plugin-marketplace triggers are mixed into the same README as install triggers. environment_fit scores 2: per-platform paths, no-admin operation, and an offline sandbox are documented.

information_architecture scores 2: docs/install is layered, but three overlapping documents (MASTER SPEC, HANDOVER, MISSION) cover the same ground with redundancy. install_notes scores 2: INSTALLATION.md gives a full directory layout, CLI commands, and exit codes. naming_stability scores 1: the product is variously CC Director, DevThrottle, cc-director and cc-launcher, and binary names disagree with doc names. examples_and_faq scores 1: install examples and an offline walkthrough exist, but there is no FAQ and examples are asserted without outputs. known_limitations scores 2: MISSION and HANDOVER plainly list unverified items and known defects. license scores 3: a complete MIT text with a clear copyright holder. versioning_changelog scores 1: no CHANGELOG, inconsistent version numbers across docs, and only release-manifest mentions per-component versions. maintenance_responsibility scores 1: SECURITY.md gives response timelines, but the publisher is unverified and ownership is unclear.

output_usability scores 1: README describes capabilities, but no runnable artifact or UI evidence appears in the repository, so output usability cannot be confirmed from source. marginal_value scores 1: the orchestration value proposition is clear, but the repository is mostly docs and a plugin marketplace with no core product code in evidence. cost_benefit scores 1: a five-minute, no-account install is claimed, yet the HANDOVER shows an install can leave a machine un-installable, so the cost/benefit carries visible risk without quantification.

claim_traceability scores 1: most capability claims point to the external devthrottle.com site with no in-repo code or tests to trace. cross_source_corroboration scores 1: HANDOVER and MISSION corroborate each other on one incident, but README marketing claims conflict with the defects recorded in the docs. fact_inference_separation scores 2: HANDOVER explicitly separates proven, unproven, and inferred findings, and MISSION flags unverified items.