Area: use-cases
What happened?
use-cases/claude-code-plugin/skills/memory-tools.md has alwaysInclude: true, so its guidance is injected into every plugin session. Under "## Available Tools" it lists:
search_memories
get_memory ("Retrieve full details of a specific memory by ID")
Neither tool is implemented anywhere in the plugin slice. The MCP server (mcp/server.js, line 15) and commands/ask.md (line 18) expose exactly one tool: evermem_search (params: query required, limit default 10 / max 20). There is no get-by-id capability at all, so get_memory describes functionality that does not exist. The result is that the always-on instructions tell the agent to call tools the server will reject, and never mention the tool that actually works.
Steps to reproduce
- Inspect the always-injected skill:
grep -n 'search_memories\\get_memory' use-cases/claude-code-plugin/skills/memory-tools.md -> matches on lines 12-13.
- Inspect the actual MCP tool surface:
grep -rn 'name:' use-cases/claude-code-plugin/mcp/server.js -> only evermem_search.
- Confirm no implementation of the named tools:
grep -rn 'search_memories\\get_memory' use-cases/claude-code-plugin/ -> only the skill doc; no server handler, no command.
- In a plugin session, the agent (following the skill) attempts to call
search_memories / get_memory; the MCP server has no such tools, so the calls cannot succeed, while evermem_search is never surfaced.
Expected vs actual
- Expected: the always-injected skill names the tool the MCP server actually exposes (
evermem_search, params query/limit), matching mcp/server.js and commands/ask.md.
- Actual: it names two non-existent tools (
search_memories, get_memory) and omits evermem_search entirely; get_memory's get-by-id behavior does not exist anywhere.
Environment
- Repo: EverMind-AI/EverOS
- Slice: use-cases/claude-code-plugin
- Files: skills/memory-tools.md, mcp/server.js, commands/ask.md
- Commit: 306dcfe (upstream/main as filed)
- Node: (any, server.js is plain Node)
- OS: any
Logs or screenshots
$ grep -n 'search_memories\\get_memory' use-cases/claude-code-plugin/skills/memory-tools.md
12:- **search_memories**: Search past conversations using semantic + keyword matching
13:- **get_memory**: Retrieve full details of a specific memory by ID
$ grep -n 'name:' use-cases/claude-code-plugin/mcp/server.js | head
15: name: 'evermem_search',
128: name: 'evermem',
151: case 'evermem_search':
A fix is staged at https://github.com/Fearvox/EverOS/pull/new/proofstorm/r1-doc-dead-xrefs.
Area: use-cases
What happened?
use-cases/claude-code-plugin/skills/memory-tools.mdhasalwaysInclude: true, so its guidance is injected into every plugin session. Under "## Available Tools" it lists:search_memoriesget_memory("Retrieve full details of a specific memory by ID")Neither tool is implemented anywhere in the plugin slice. The MCP server (
mcp/server.js, line 15) andcommands/ask.md(line 18) expose exactly one tool:evermem_search(params:queryrequired,limitdefault 10 / max 20). There is no get-by-id capability at all, soget_memorydescribes functionality that does not exist. The result is that the always-on instructions tell the agent to call tools the server will reject, and never mention the tool that actually works.Steps to reproduce
grep -n 'search_memories\\get_memory' use-cases/claude-code-plugin/skills/memory-tools.md-> matches on lines 12-13.grep -rn 'name:' use-cases/claude-code-plugin/mcp/server.js-> onlyevermem_search.grep -rn 'search_memories\\get_memory' use-cases/claude-code-plugin/-> only the skill doc; no server handler, no command.search_memories/get_memory; the MCP server has no such tools, so the calls cannot succeed, whileevermem_searchis never surfaced.Expected vs actual
evermem_search, paramsquery/limit), matchingmcp/server.jsandcommands/ask.md.search_memories,get_memory) and omitsevermem_searchentirely;get_memory's get-by-id behavior does not exist anywhere.Environment
Logs or screenshots
A fix is staged at https://github.com/Fearvox/EverOS/pull/new/proofstorm/r1-doc-dead-xrefs.