Skill

semantic-recall

Query Pinecone (or fallback to grep-over-markdown) for context relevant to a task. Returns ranked snippets with cited paths. Used by suggest-companies, meeting-prep, weekly-plan, self-healing-context when CONTEXT_STORE=pinecone. Caller decides what to do with results.

npx claudepluginhub danielrosehill/claude-code-plugins --plugin career

Tool Access

This skill is limited to using the following tools:

ReadBash(grep *)Bash(rg *)

Preview

Pulls relevant context from the workspace for the current task. Two backends:

SKILL.md

Similar Skills

cache-components

139.3k

Guides Next.js Cache Components and Partial Prerendering (PPR): 'use cache' directives, cacheLife(), cacheTag(), revalidateTag() for caching, invalidation, static/dynamic optimization. Auto-activates on cacheComponents: true.

cache-components

mcp-builder

124.2k

Guides building MCP servers enabling LLMs to interact with external services via tools. Covers best practices, TypeScript/Node (MCP SDK), Python (FastMCP).

9 files

anthropics-skills-13

Stats

Stars0

Forks0

Last CommitApr 28, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

Semantic Recall

Pulls relevant context from the workspace for the current task. Two backends:

CONTEXT_STORE=pinecone → Pinecone semantic search.
CONTEXT_STORE=markdown (default) → ripgrep over working-folder.

The skill returns citations — paths + chunks. The caller is responsible for using them.

Inputs

$ARGUMENTS:

Required: --query="<text>" — natural-language query.
Optional: --type=<ground-truth|crm|domain-note|recommendation|brief|plan|meeting|inbox|all> — filter by chunk type. Default: all.
Optional: --top=<n> — top-N results. Default: 8.
Optional: --min-score=<float> — min relevance score (Pinecone only). Default: 0.6.

Examples

/* invoked internally by other skills, not via slash command */
semantic-recall --query="prior outreach to AI policy companies in Tel Aviv" --type=crm --top=10
semantic-recall --query="user's stated salary floor and reasoning" --type=ground-truth

Procedure

1. Load config

CONTEXT_STORE. If pinecone, also PINECONE_INDEX.

2a. Pinecone backend

Use Pinecone MCP search-records: query text, filter by type if provided, top-K = --top.
Filter results below --min-score.
For each result, return: {path, section, score, snippet (from metadata), type}.

2b. Markdown backend (fallback)

rg -n over working-folder/ matching files for --type (e.g. --type=crm → crm/*.md).
Score by term-frequency match on query keywords (cheap heuristic).
Return same shape minus score (use match-count instead).

3. Return

Print machine-parseable lines:

result: path=<path> section="<heading>" score=<float> type=<type>
snippet: <one-line preview>

…or a JSON block if caller passed --json.

Guardrails

Always cite the path. A recall result without a path is unusable downstream.
Don't summarise the snippet — return raw text so the caller can decide.
If CONTEXT_STORE=pinecone but the manifest is older than 24 hours, warn that recall may be stale.

Failure modes

Pinecone MCP unavailable → automatically fall back to markdown backend with a one-line note.
Empty workspace → return zero results, not an error.