halyard

Route questions to human experts via Slack and capture answers as organizational knowledge. Use when stuck, blocked, need human input, facing ambiguity, making design decisions with multiple valid approaches, need domain expertise not in the codebase, want to check with someone, or need verification before significant effort. Triggers on 'ask someone', 'get help', 'need an expert', 'check with the team', 'I'm stuck', 'get a second opinion', 'who should I ask'. Always summarize helpful answers to build organizational knowledge.

Gather background context before starting work or answering questions about a topic. Use when the user asks 'what do we know about', 'background on', 'catch me up on', 'context on', 'research X before starting', 'what's the history of', 'look into', 'investigate', 'find out about', or needs orientation on an unfamiliar area. Also use proactively at the start of a task when the work touches a named system or feature area (auth, payments, ingestion, a specific service), requires architectural judgment, or the codebase has prior decisions that could constrain the approach. Skip for mechanical edits, single-file changes, or clear-cut bug fixes where prior context is unlikely to change behavior. Combines knowledge base search, codebase exploration, and web search.

Log completed work to the team knowledge base. Use when finishing a session, completing a task, wrapping up, making a decision, or when prompted to record, document, save, or summarize what was done. Triggers on 'log this', 'write this up', 'save to knowledge base', 'document what we did', 'record this decision', 'summarize this session'. Captures work summaries, architectural decisions, process documentation, and learnings for team visibility.

Mark definitional terms when authoring knowledge — link org/product/industry jargon to its canonical DEFINITION entry so the graph connects. Use when writing or updating a knowledge entry (summarize_work, summarize_conversation, upsert_knowledge) and the text leans on a named concept, product surface, acronym, or coined phrase a newcomer or agent wouldn't already know. Triggers on 'what does X mean', 'define this term', 'add a glossary entry', 'link the term', or first use of load-bearing jargon in an entry you're writing.

Query the team knowledge base for recent activity, past decisions, and work context. Use when the user asks what was done, what happened, what changed, wants a standup summary, daily update, weekly review, needs to catch up, get up to speed, understand previous work, or asks about past decisions. Triggers on 'show me recent work', 'what did the team do', 'review activity', 'what's been going on', 'standup', 'what was accomplished'. Searches work summaries, expert Q&A, and decision records.

Review and clear the knowledge base inbox — the queue of AI-generated candidate entries awaiting human approval before they go live. Use when the user asks to review the inbox, triage pending knowledge, approve or reject candidate entries, clear the review queue, or check what's waiting to be filed. Triggers on 'review the inbox', 'triage knowledge', 'what's pending approval', 'clear the queue', 'accept/dismiss candidates', 'review draft entries'. Candidates accepted here enter the live, searchable knowledge base.