From neohive
Use when the user says "save what I learned", "capture this session", "extract learnings", or at the end of any session (also fires automatically from the stop hook). Scans the conversation transcript, extracts corrections, conventions, decisions, and gotchas worth persisting, deduplicates against NeoHive, stores up to 5 new memories, and audits CLAUDE.md for instruction gaps.
How this skill is triggered — by the user, by Claude, or both
Slash command
/neohive:capture-session-learningsThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You capture valuable knowledge from the current session into NeoHive semantic memory so future sessions inherit it. This skill is the end-of-session counterpart to `/neohive:load-context`.
You capture valuable knowledge from the current session into NeoHive semantic memory so future sessions inherit it. This skill is the end-of-session counterpart to /neohive:load-context.
Analyze the conversation transcript of this session. Extract learnings worth persisting across sessions and store them in the vector memory system via MCP tools. Also check whether the memory usage instructions in ~/CLAUDE.md need improvement.
Scan the full session transcript for these categories of knowledge:
error_pattern (importance 7-8)convention or directive (importance 8-9)decision (importance 6-8)insight (importance 6-7)error_pattern (importance 7)idiom (importance 6-7)syntax_rule or stdlib_reference (importance 7-9)Be selective. Only extract knowledge that would be valuable in a future session on a different day. Skip:
For each candidate, formulate a clear, self-contained description. It should make sense to someone (or an LLM) reading it without the context of this session.
For each candidate learning:
memory_recall with a semantic query that describes the learning (use memory_recall here, not memory_context — deduplication needs a narrow single-type search against the specific learning, not a broad context load)This is the critical step. The deduplication heuristic is: if the memory system already had this insight, Claude should have found it during the session. If Claude didn't find it, either the insight is new, or the existing memory's embedding doesn't cover this query angle — either way, storing a new entry is correct.
For each learning that passed deduplication, call memory_store with:
content: Clear, self-contained description of the learningtype: The appropriate type from Phase 1tags: 3-6 relevant tags for filteringimportance: As suggested in Phase 1metadata: Include { "source": "capture-session-learnings", "session": "${CLAUDE_SESSION_ID}" }Check whether this session revealed a gap in how Claude uses the memory system:
memory_context at session start?If you identify a gap, read ~/CLAUDE.md and check the ## Cognitive Memory System section. If the instructions don't already cover this gap, update that section with a specific, actionable instruction. Use the Edit tool — do not rewrite the entire section, just add or refine the specific instruction.
After completing all phases, output a brief summary:
Memory revision complete:
- Candidates found: N
- Already known (skipped): N
- New memories stored: N (list IDs)
- CLAUDE.md updated: yes/no (what changed)
Promise.allSettled instead of Promise.all when partial failures are acceptable in our batch processor" is better than "Handle promise errors properly".npx claudepluginhub neohiveai/neohiveclaude --plugin neohiveFetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
Applies a firm's KYC/AML rules grid to parsed onboarding records: assigns risk rating, checks required documents, outputs rule outcomes with citations, and routes for escalation.
Generates daily or weekly digests of activity from connected sources (chat, email, docs, tasks, CRM), highlighting action items, decisions, mentions, and project updates.