AI Ask & End-to-End Product Workflows (ah ask, ah research, ah spec, ah plan, ah implement)
Agent Harbor connects your CLI directly to online web LLM services (ChatGPT Pro, OpenAI Deep Research, Gemini Deep Research, Claude Web, Grok) and provides a uniform phase workflow (new, refine, consolidate, status) that can take a single product prompt from feature discovery all the way to a verified multi-milestone implementation.
Commands
| Command | Artifact & Convergence Goal |
|---|---|
ah ask [OPTIONS] [QUESTION] | Queries browser-automated web LLM services (--service chatgpt|claude|gemini|grok, --all) with --deep-research, --deep-think, --think, and --attach-bundle <name>. |
ah bundle <NAME> | --list | Packs the files and primers declared under [bundles.<name>] into a reusable context bundle for ah ask, ah research, and ah spec. |
ah research [new|refine|consolidate|status] | Feature Map Discovery & Completeness: Fans out deep-research queries across facets and providers to build and refine a structured feature catalog (Feature-Index.org + facet files). |
ah spec [new|refine|consolidate|status] | Specification Soundness: Synthesizes and refines engineering specifications using Pro / Deep-Think multi-model critique loops until soundness converges. |
ah plan [new|refine|consolidate|status] | Milestone-Graph Coherence: Decomposes specifications into a dependency-ordered milestone graph and refines it to coherence. |
ah implement [OPTIONS] [<plan-or-milestone>] | Milestone Completion: Drives multi-milestone implementation and adversarial verification inside isolated copy-on-write workspaces. |
ah review [OPTIONS] | Fans out parallel adversarial code and spec-compliance reviews across multiple models in cheap copy-on-write workspace snapshots. |
Example: Querying Online Web LLMs (ah ask)
# Ask ChatGPT Pro / Claude / Gemini / Grok with maximum reasoning depth
ah ask --deep-think --attach-bundle commercial-model "Review the seat-charge model for contradictions"
# Run multi-provider Deep Research from the CLI
ah ask --deep-research --all --topic "quic-nat-traversal" --output-dir ./research/Example: From a Single Prompt to a Complete Product
# 1. Discover and refine a comprehensive feature map using Deep Research
ah research "Build a distributed time-series telemetry service with SQL and PromQL query engines"
ah research refine --facets market,technical,competitive,user
# 2. Author and refine the engineering specifications with Pro / Deep-Think models
ah spec new
ah spec refine --until-converged
# 3. Generate and refine the multi-milestone execution plan
ah plan new
ah plan refine --until-converged
# 4. Execute the multi-milestone implementation and parallel cross-vendor reviews
ah implement
ah review