Skip to content

CHANGELOG.md

Latest commit

 

History

History
106 lines (78 loc) · 6.67 KB

CHANGELOG.md

File metadata and controls

106 lines (78 loc) · 6.67 KB
Raw

Changelog

[1.0.3] - 2026-05-31

Added

  • Added new install-skill CLI command to easily install the bundled knowledge-builder skill directly into an agent's skills directory.
  • Enabled install-skill to automatically default to the current working directory (.) if no path argument is provided.
  • Added comprehensive integration tests in tests/cli.test.ts to verify programmatic copying and overwriting of bundled skills.

[1.0.2] - 2026-05-31

Refactored

  • Migrated MCP server setup from the deprecated Server to the recommended McpServer API (@modelcontextprotocol/sdk/server/mcp.js).
  • Migrated tool registrations from manual setRequestHandler + ListToolsRequestSchema/CallToolRequestSchema to server.registerTool.
  • Replaced plain JSON Schema objects with Zod-based type-safe input schemas (z.object()) for all tools — enables automatic argument validation by the SDK.
  • Added zod as a runtime dependency.

Fixed

  • Fixed run_workflow inputs parameter type: changed from string (requiring manual JSON.stringify by the caller) to object (z.record(z.string(), z.string())), matching MCP conventions. Updated toolRunWorkflow to accept unknown and coerce values to string instead of calling JSON.parse.
  • Fixed list_docs tool description which had been incorrectly set to a context-gatherer agent description unrelated to the tool's purpose.

Changed

  • Added MCP server instructions field with guidance on when and how to use each tool, recommended call order (search_docs → get_manifest → get_doc/get_doc_section/get_resource → run_workflow), and the two knowledge formats.
  • Improved all tool descriptions with usage guidance, cross-tool hints, and path format examples:
    • search_docs: clarified it should be called first before any task; documented scoring order.
    • get_doc: added hint to prefer get_doc_section for large documents.
    • get_doc_section: noted context efficiency and partial heading match support.
    • get_resource: added "use get_manifest first to discover paths" and path format example.
    • get_manifest: added "always call before run_workflow" guidance.
    • run_workflow: noted side effects (file creation, script execution).
  • Improved all tool parameter descriptions with path format clarifications (relative to knowledge directory root) and concrete examples.

[1.0.1] - 2026-05-30

Security

  • Fixed path traversal vulnerability in get_doc, get_manifest, and run_workflow — all tool inputs are now validated to stay inside the knowledge directory
  • Fixed command injection in workflow run steps — replaced execSync string concatenation with spawnSync array args (shell is never invoked)
  • Fixed command injection in CLI git operations (add, update) — replaced execSync string interpolation with spawnSync array args
  • Fixed PowerShell command injection in export on Windows — single quotes in paths are now escaped with ''
  • Fixed symlink traversal in get_resource, get_doc, get_manifest, run_workflow — realpathSync is used to resolve symlinks before boundary check, and the resolved canonical path is returned to callers
  • Fixed script path traversal in workflow run steps — script path is validated to stay inside the knowledge unit folder before execution
  • Fixed missing cleanup of temporary clone directory on add failure
  • Added interpreter whitelist for workflow scripts (python, python3, node, pwsh, powershell, bash, sh)

Performance

  • walkDir in both MCP server and CLI now skips .git, .cache, and node_modules directories

Fixed

  • Frontmatter parsing in MCP server now uses \r?\n split for consistent behavior on Windows (CRLF line endings)

Changed

  • Added repository, homepage, and bugs fields to package.json for npm registry display

[1.0.0] - 2025-05-24

Initial release of Knowledge Shelf — a personal knowledge management system for AI coding assistants, built as an MCP server.

MCP Server

Give your AI on-demand access to your local knowledge base through 7 tools:

  • list_docs — Discover all available knowledge units and documents
  • search_docs — Find relevant knowledge by keyword with relevance scoring
  • get_doc — Read the main documentation of a knowledge unit
  • get_doc_section — Read a specific section without loading the full document
  • get_resource — Read code files, templates, and configs from a unit
  • get_manifest — Inspect what a unit contains (modules, workflows, references)
  • run_workflow — Execute multi-step automation (scaffold, copy files, run scripts)

Knowledge formats supported:

  • Manifest-based units — folders with manifest.json for code patterns, templates, and workflows (supports modules, scripts, references, placeholders)
  • Standalone .md documents — plain markdown files with optional YAML frontmatter (description, tags, aliases)

Search scoring: aliases (12) → title (10) → tags (8) → type (7) → modules/filename (6) → sections (4) → description (2)

Security: Path traversal protection on get_resource

Default knowledge directory: ~/.knowledge

Knowledge Shelf CLI

Manage knowledge sources from git repositories — like npm for documentation.

  • init — Initialize knowledge directory with .gitignore and registry
  • add — Clone from git repos; supports --path, --name, --all, and sparse checkout
  • list — Show installed sources with origin and version
  • info — Detailed view of a knowledge unit (type, modules, size, source)
  • search — CLI keyword search using the same scoring as the MCP tool
  • validate — Check manifest integrity (required fields, file references, workflow JSON)
  • update — Pull latest from remotes (respects pinned sources)
  • pin / unpin — Version locking to prevent unwanted updates
  • export — Package a unit as .zip or .tar.gz for sharing
  • remove — Delete a knowledge unit and its registry entry

Directory resolution order: --dir flag → KNOWLEDGE_DIR env → walk-up .registry.json → ~/.knowledge

Bundled Skill

  • knowledge-builder — AI-assisted skill for scaffolding new knowledge units, generating reference docs from source code, converting loose docs into manifest-based units, and validating structure

Documentation

  • docs/cli-reference.md — Full CLI command reference (11 commands, options, exit codes)
  • docs/tools-reference.md — MCP tool parameters, responses, and examples
  • docs/knowledge-folder-guide.md — How to organize knowledge directories and write manifests
  • docs/publishing-knowledge.md — Git repo layouts for team sharing