Skip to content

TODO.md

Latest commit

 

History

History
90 lines (76 loc) · 6.86 KB

TODO.md

File metadata and controls

90 lines (76 loc) · 6.86 KB

Claude Code Web Hooks - TODO

Last Updated: 2026-04-10

✅ Completed

  • Chosen project name: Claude Code Web Hooks
  • Created standalone project folder for the project scaffold
  • Created initial design.md
  • Created initial changelog.md
  • Created initial TODO.md
  • Added project-owned hook source files under hooks/
  • Added settings.example.json for Claude Code integration
  • Added installation/how-to-add guidance into design.md

📋 Tasks To Do

Documentation / Planning

  • Refine standalone project scope and installation model
  • Decide final file layout for hooks/, src/, and install/
  • Define configuration strategy for user-scope vs project-scope installation

Extraction / Implementation

  • Extract current WebSearch hook into project-owned implementation files
  • Extract current WebFetch scraper hook into project-owned implementation files
  • Move shared helper logic into reusable modules
  • Add installer/update script for safe settings.json merge
  • Add uninstaller for project-owned hook removal

Behavior / Quality

  • Improve API key pool format to avoid delimiter ambiguity
  • Reduce noisy hook messaging while preserving decision visibility
  • Make parallel mode return all successful provider results and report partial failures
  • Add domain heuristics for better WebFetch scrape escalation decisions
  • Refine WebFetch low-text structured portal detection so template-heavy pages with repeated metadata blocks do not fall through as fetch-readable
  • Add tests or repeatable verification flow for hook decision branches

Multi-provider roadmap

  • Add search provider abstraction layer
  • Add Tavily Search adapter
  • Add search provider policy config (fallback / parallel)
  • Wire WebSearch flow to use provider policy instead of hardcoded provider-specific routing
  • Normalize Tavily Search responses into the project’s shared search-result format
  • Extend verify coverage for multi-provider search behavior
  • Research and evaluate Exa.ai as an additional search-provider adapter
  • Decide whether Exa should be search-only first or enter a later extract-provider phase
  • Design a bounded WebFetch extractor-provider slice for Tavily Extract and Exa Contents while keeping WebSearchAPI.ai Scrape as the active backend
  • Decide that future WebFetch extractor support should use a provider-policy model with ordered fallback
  • Implement three interchangeable WebFetch extraction backends from the first rollout (WebSearchAPI.ai Scrape, Tavily Extract, Exa Contents)
  • Finalize three-backend WebFetch verification and release-sync wording
  • Run live smoke tests with real provider keys for WebSearchAPI.ai Scrape, Tavily Extract, and Exa Contents
  • Design target-aware install / uninstall / verify support for multiple runtime targets (claude-code, copilot-vscode, copilot-cli, all)
  • Define Copilot-on-VS-Code compatibility wrappers / config placement without breaking the current Claude Code path
  • Extend the Copilot compatibility layer to also support Copilot CLI payload/input-output rules
  • Define multiple-target installer semantics that keep the design open to future targets beyond claude-code and copilot-vscode
  • Add explicit non-blocking coexistence support for CCS MCP WebSearch (mcp__ccs-websearch__WebSearch) without taking ownership of the MCP tool path
  • Extend CCS MCP coexistence so the original CCS MCP result and a claude-code-web-hooks companion result can both be surfaced together via PostToolUse
  • Add CCS MCP failure fallback so failed MCP runs can still surface provider-backed fallback context via PostToolUseFailure
  • Verify native WebSearch substitution and CCS MCP pass-through together without double-search behavior
  • Verify CCS MCP companion replacement behavior with preserved original MCP output plus appended companion results
  • Verify CCS MCP failure fallback behavior with provider-backed additional context

Release readiness checklist

  • Add .gitignore
  • Add LICENSE
  • Add public-facing README.md with install and use instructions
  • Add Claude Code flow diagram to README
  • Add fixtures directory for verification
  • Add verify.sh
  • Review repository text for any remaining machine-specific paths that should stay placeholder/generic in public release
  • Decide whether to keep or remove direct host heuristics before public release
  • Create repository and push when ready
  • Create first tagged release when ready

📜 History

Date Changes
2026-04-10 Reworked the CCS MCP failure fallback block into a result-first layout so the fallback search result appears before the preserved raw CCS error, extended verification to assert the new summary text and section order, and synced the governed docs/TODO wording to the new visibility contract.
2026-04-05 Extended CCS MCP coexistence so failed MCP runs now also trigger repo fallback context: websearch-mcp-companion.cjs now handles PostToolUseFailure, install/settings examples now include the failure-side matcher, verification covers both success-side dual output and failure-side fallback context, and the governed docs/phase/patch set was synced to that bounded contract.
2026-04-04 Extended CCS MCP coexistence from allow-only pass-through into a dual-output model: added a PostToolUse companion hook that preserves the original CCS MCP result and appends a claude-code-web-hooks companion result via updatedMCPToolOutput, then synced install/uninstall/settings/verify/docs/phase artifacts to the new contract.
2026-03-28 Implemented and verified the narrow WebFetch heuristic refinement for low-text structured portal pages, then completed the three-backend WebFetch extraction rollout in repo state: added interchangeable extraction backends for WebSearchAPI.ai Scrape, Tavily Extract, and Exa Contents; added one-active-backend-per-request selection with ordered fallback; updated install/uninstall/settings/verify/docs/phase files; completed real-key smoke testing for Tavily Extract and Exa Contents plus ordered fallback behavior; and staged then implemented target-aware install / uninstall / verify support for the multiple-target model (claude-code, copilot-vscode, copilot-cli, all) including Copilot compatibility wrappers and Copilot CLI repo-hook support.
2026-03-27 Audited design.md and phase/ against the current implementation, aligned wording to the active provider set (WebSearchAPI.ai, Tavily, Exa), normalized phase titles to 001-005, and closed stale TODO items that had already been implemented.
2026-03-20 Created new standalone project scaffold claude-code-web-hooks with initial design, changelog, and TODO to separate web hook logic from external gateway/runtime ownership.