Skip to main content

Roadmap

Missing pieces​

  • Command contract: an explicit CTS-style contract for its public commands
  • Exit code table documenting mapping to CTS bands
  • JSON output envelope schema (CommandOutput.schema.json), or documented adherence to one
  • --help output and documented CLI usage examples
  • Version output and a release checklist
  • Structured error examples and a documented machine-readable error object
  • Build/release artifacts and release verification (build_verified=false, release_verified=false)
  • Local config.toml setup documented as a required bootstrapping step (copy from config.example.toml, set the API key)

Next steps​

  1. Document a CTS-compliant command contract (purpose, invocation, inputs, stdout/stderr behavior, machine-readable mode, exit codes, examples, stability notes)
  2. Publish an exit code table and ensure Summarizer.py exits match it
  3. Add a documented JSON envelope (or CommandOutput.schema.json) and a --json flag that emits it without progress text on stdout
  4. Add --help and --version flags, with runnable examples for both automation and human use
  5. Create a release checklist and produce build artifacts to move verification.build_verified to true
  6. Document bootstrap steps: copy config.example.toml, set the environment API key, validate model endpoint connectivity
  7. Add lightweight tests validating ModelClient fallback behavior (Ollama → OpenAI) and end-to-end summary generation for one sample project

Potential improvements​

  • Emit a strict CTS JSON envelope by default when --json is passed, keeping human-readable output separate
  • Expose progress and verbose logs exclusively on stderr so stdout stays machine-readable
  • Add a --dry-run mode for any future mutating operations
  • Cache model responses with a local replay mode, to avoid repeated API calls during iterative development
  • A small CLI shim that validates ProjectIndex.md counts and reports mismatches before a full (paid) run

Creative enhancements (ideas on file, not committed)​

  • Delta-driven re-evaluation: track file hashes per project and only re-run analysis for changed projects, saving model calls
  • A local model-benchmark harness comparing multiple LLMs (Ollama vs. OpenAI) on a held-out project set, surfacing cost/quality tradeoffs
  • DuckDB export of summaries for ad-hoc SQL analytics and historical trend queries across portfolio snapshots
  • An automated migration assistant suggesting manifest fixes or missing fields for child projects, based on template comparisons
  • Webhook/notification integration to push top-priority project changes to a private chat or task board