README.md

Documentation Index

This docs index helps you navigate architecture notes and diagrams.

• ADR-0001: Minimal Agent CLI — design context, decisions, and contracts.

  • Link: docs/adr/0001-minimal-agent-cli.md

• ADR-0002: Unrestricted toolbelt (files + network) — risks, contracts, and guidance.

  • Link: docs/adr/0002-unrestricted-toolbelt.md

• ADR-0003: Toolchain & Lint Policy (Go + golangci-lint) — pin Go via go.mod and a known-good golangci-lint; CI and local workflows align.

  • Link: docs/adr/0003-toolchain-and-lint-policy.md

• ADR-0004: Default LLM Call Policy — default temperature 1.0, capability-based omission, one-knob rule, and observability fields.

  • Link: docs/adr/0004-default-llm-policy.md

• ADR-0005: Harmony pre-processing and channel-aware output — pre-stage HTTP call, parallel read-only tools, validator/audit with stage tags, and deterministic channel routing.

  • Link: docs/adr/0005-harmony-pre-processing-and-channel-aware-output.md

• ADR-0006: Image generation tool (img_create) — minimal Images API integration, repo‑relative saves, env passthrough, and transcript hygiene.

  • Link: docs/adr/0006-image-generation-tool-img_create.md

• ADR-0010: Adopt SearXNG & network research toolbelt (CLI-only) — credible web discovery with provenance via SearXNG and a safe, CLI-only toolbelt.

  • Link: docs/adr/0010-research-tools-searxng.md

• ADR-0011: State bundle schema — versioned on-disk JSON snapshots for reproducible runs.

  • Link: docs/adr/0011-state-bundle-schema.md

• ADR-0012: Persist and refine execution state via -state-dir — file-based snapshots with scopes, restore-before-prep, and refinement controls.

  • Link: docs/adr/0012-state-dir-persistence.md

• Sequence diagrams: agent flow and toolbelt interactions.

  • Link: docs/diagrams/agentcli-seq.md
  • Link: docs/diagrams/toolbelt-seq.md
  • Link: docs/diagrams/harmony-prep-seq.md
  • Link: docs/diagrams/research-pipeline.md
  • Link: docs/diagrams/prestage_flow.md

• Architecture: Module boundaries and allowed imports between layers.

  • Link: docs/architecture/module-boundaries.md

• Tools manifest reference: precise tools.json schema and mapping to OpenAI tools.

  • Link: docs/reference/tools-manifest.md

• CLI reference: complete flag list, env precedence, exit codes.

  • Link: docs/reference/cli-reference.md

• Tool reference: Image generation tool (img_create).

  • Link: docs/reference/img_create.md

• Tool reference: HTTP fetch (http_fetch).

  • Link: docs/reference/http_fetch.md

• Tool reference: SearXNG search (searxng_search).

  • Link: docs/reference/searxng_search.md

• Tool reference: Crossref search (crossref_search).

  • Link: docs/reference/crossref_search.md

• Tool reference: PDF extract (pdf_extract).

  • Link: docs/reference/pdf_extract.md

• Tool reference: Wayback lookup (wayback_lookup).

  • Link: docs/reference/wayback_lookup.md

• Security: Threat model and trust boundaries.

  • Link: docs/security/threat-model.md

• Runbooks: Troubleshooting common errors and fixes.

  • Link: docs/runbooks/troubleshooting.md

• Migrations: Tools layout (legacy → canonical tools/cmd/* + tools/bin/*).

  • Link: docs/migrations/tools-layout.md

Additional guides will be added here as they are created.

Model parameter compatibility

Some reasoning-oriented models may not accept sampling parameters. The agent omits temperature automatically for such models while keeping the default of 1.0 for compatible families (e.g., GPT-5 variants). This avoids API errors and preserves expected defaults where applicable.