Search broke when projects outgrew one tool
Modern delivery stacks scatter truth: tasks on the board, decisions in wiki, contracts in documents, burn in finance, names in the people directory. Classic search boxes query one module at a time. Leaders ask "where is the payments integration decision?" and spend twenty minutes opening wrong tabs.
AI smart project search in WKFGo centers on smart_search—a permission-aware query that returns candidates across tasks, people, documentation, and related entities in one response. Assistants via MCP use the same tool; the chatbot can ground answers in hits instead of inventing context.
Problem: search without governance
Ungoverned search creates two failures:
- Under-find — Users give up and re-ask in Slack
- Over-expose — Bots query with superuser tokens and summarize restricted finance tasks in a public channel
Smart search inherits feature access like the UI. If a contractor cannot open a document in WKFGo, search should not surface its body to them via an assistant workaround.
Framework: intent → query → disambiguate → act
- State intent — "Find release 2.4 blockers owned by Team Payments"
- Call smart_search — Natural language or keyword; include project scope when known
- Disambiguate — Tool may return candidates when names collide; pick exact project/task
- Act — Open task, update wiki, or chain to
get_task,get_wiki_page,search_docs
AI adds value by reformulating vague questions and summarizing hit lists with links—not by skipping disambiguation.
Example intents
| User question | Search focus |
|---|---|
| "Who owns auth migration?" | People + tasks labeled auth |
| "Last decision on pricing API" | Wiki + meeting pages |
| "Overdue items for Ali" | Tasks + assignee filter |
| "Document for vendor X contract" | Documents + finance links |
Solution: smart search in daily workflows
Manager morning orientation
Replace tab hopping with one smart_search before standup:
- Blockers mentioning "production"
- Tasks stuck in review
- Wiki pages not updated in 90 days for critical subsystems
Post top hits to team channel with task links—human validates before broadcasting.
On-call incident triage
When alerts fire, search runbooks and prior incident tasks:
smart_searchsubsystem name + "incident"- Open wiki runbook via
get_wiki_page - Link new task to prior post-mortem page
Reduces repeated outages from missing institutional memory.
Developer context via MCP
IDE agents call smart_search before editing unfamiliar modules—pulling coding standards wiki, related open tasks, and recent decisions. Pair with get_context_pack for focused task work.
Executive ad-hoc questions
CEO asks "what is status of EU compliance work?" — search across projects; if multiple matches, disambiguate portfolio vs program name before briefing.
Tips for better hits
- Include project name when portfolio is large
- Prefer concrete nouns (service names, release tags) over "that bug"
- Refresh stale wiki—search quality equals documentation quality
- Do not cache sensitive hit lists in public logs
Integration with MCP and daily drivers
smart_search sits early in many MCP prompt chains. A developer might search before get_context_pack; an executive might search before executive_brief. The pattern is consistent: disambiguate entity names (project and task titles collide in large portfolios), then drill into get_task, get_wiki_page, or search_docs for full bodies.
For my_day and my_queue, search complements personal queues—you lookup historical context ("when did we last touch compliance?") while queues show today's commitment. Teams running bilingual programs should search using the language indexed in wiki titles and task descriptions; mixed-language portfolios benefit from trying both Farsi and English keywords when hits are sparse.
Document your team's search cheat sheet in wiki: common release codenames, subsystem nicknames, and finance label conventions. Search quality rises when vocabulary is shared—not when everyone invents a different name for the same service.
Anti-patterns
- Using admin tokens in shared bots — bypasses governance
- Searching instead of maintaining board hygiene — search finds chaos too
- Ignoring candidate disambiguation — wrong task updates hurt
- Expecting search to replace
decision_log— decisions still need explicit logging
Search fails when tasks, wiki, and decisions live in separate silos. smart_search spans permission-scoped entities—train teams to search before asking in channel. AI answers should cite entity type and ID from results, not paraphrase from memory.
Run a monthly audit: do AI answers cite task IDs from MCP? If not, tighten prompts to require citations before any stakeholder-facing draft leaves the team.
FAQ
How is smart_search different from chatbot Q&A?
Search returns ranked hits; Q&A synthesizes answers—often after search or search_docs.
Cross-project search?
Yes within permissions; large portfolios should scope by project when possible.
Bilingual queries?
Teams work in English, Farsi, or mixed; use the language stakeholders indexed content in.
API access?
MCP exposes smart_search for assistants; see /api-docs for REST patterns.
Find project truth in one query
Stop bookmark archaeology. Search once, disambiguate, act with permission-aware hits.
Try it now
Put these patterns on live project data—not slide decks.