Bough Work and Cost Explorer

Permissions and side effects: the README limits the tool to reading local Claude Code, Codex, and Pi history plus optional Git history; it says the server binds to 127.0.0.1, makes no network requests by default, and writes only to its own settings directory when a project is renamed. It also documents --no-repo, supporting least_privilege=3 and external_effects=3. The installer downloads and installs a binary and documents checksums, but runtime actions are not described as requiring step-by-step confirmation, so user_confirmation=2. Data flows are described in detail, but full prompts may contain sensitive information and the files show no redaction or sensitive-content protections, so sensitive_data_handling=1. The module declares only x/term and indirect x/sys; CI includes security scanning and CodeQL, but no dependency audit results are supplied, so dependency_security=2. Repository and agent directories are read-only and the rename file location is documented, but no restore or undo process is described, so rollback=2. The license names a copyright holder, but there is no verified publisher identity or broader evidence of maintainer attribution, so source_attribution=2.

The README, SECURITY file, and CI configuration are broadly consistent about local read-only behavior, loopback serving, and no data transmission. CI declares cross-platform builds, tests, static analysis, and fuzzing; this review does not treat those declarations as executed verification, so self_consistency=2. The Go module has few dependencies, while embedded pages and offline rates reduce runtime dependencies; building still requires Go dependency retrieval and installation channels rely on release hosting, so dependency_availability=2. The docs mention NA for unpriced models and continuing when Git history is unavailable, but give little detail about user-visible parse or permission errors, so failure_messages=1.

The intended audience and use are clear: individual developers using Claude Code, Codex CLI, or Pi who want project and task views of time, prompts, and estimated cost. The README covers GUI, text, JSON, and filtering use cases, so audience_and_scenarios=3. Boundaries are concrete: cost is not a bill, unsupported rates produce NA, only three agents are covered, and grouping thresholds are fitted to one person's history, so capability_boundaries=3. CLI options cover project selection, output modes, agent filtering, and disabling Git reads, but automatic discovery triggers and error paths are not fully specified, so trigger_precision=2. Install methods and CI cover macOS, Linux, and Windows, though some installation paths use shell or PowerShell and require network access, so environment_fit=2.

The README is organized around usage, features, limitations, formats, layout, and contribution, so information_architecture=3. Install guidance includes platform commands, manual and Go installs, upgrade behavior, and checksum notes, so install_notes=3. Core names and terms are reasonably stable, but there is no API or compatibility policy, so naming_stability=2. The README contains several command examples and detailed feature explanations but no dedicated FAQ or extensive troubleshooting section, so examples_and_faq=2. Limitations cover estimated costs, thresholds fitted on one history, supported agents, Git count discrepancies, and early status, so known_limitations=3. The repository contains the MIT license text, so license=3. SECURITY says there are no releases yet and fixes land on main; there is no changelog or explicit supported-version table, so versioning_changelog=1. Contribution guidance, issue/PR invitations, and a security reporting path provide maintenance entry points, but maintainer responsibility is not clearly assigned, so maintenance_responsibility=2.

The visual interface, records linked to original prompts, task-level pricing, and text/JSON exports support inspection and reuse, so output_usability=3. The tool turns scattered sessions into tasks, sittings, file activity, and costs, and explains the granularity gap in existing usage tools, demonstrating a clear added use, so marginal_value=3. A single binary, small dependency set, and local processing help keep operating costs low, but results depend on history formats and thresholds fitted to one person's data, so cost_benefit=2.

The docs point to in-repository format and tuning notes and explain differences between Git commits and agent records; many functional and security claims nevertheless rely primarily on the README, so claim_traceability=2. The README, SECURITY file, module manifest, and CI provide some corroboration on boundaries and design, but no independent audit results are supplied and CI files are configuration evidence only, so cross_source_corroboration=2. The docs distinguish agent records from surviving Git commits, estimated rates from actual bills, and fitted thresholds from established facts, handling fact/inference boundaries thoroughly, so fact_inference_separation=3.