YashuLanki
@YashuLanki
Public Skills
answer-eval
by YashuLanki
Score whether Claude's ANSWERS are right, not just whether the code runs — run the generated question set through the real MCP tools and mark the results. Use before shipping a change that could affect how questions get answered, or when asked to check answer quality, citation accuracy, or whether the system makes things up.
full-sweep
by YashuLanki
Use to run a whole-system health and correctness audit of Vaulter AI -- screening correctness, report readability, confidential-data leaks, shared-folder hygiene, and whether a fresh teammate could still install it. Invoke when the user asks to "check everything", before a release that matters, after a batch of changes, or on a periodic cadence (monthly is the intent).
cleanup
by YashuLanki
Use when asked to clean up the repo — remove dead files, folders, docs, config constants, imports, or stale doc sections left behind by earlier work. Also for "what's dead in here?", "tidy this up", or after a large refactor.
vaulter-rebuild
by YashuLanki
Use when working on the Vaulter AI rebuild described in docs/REBUILD_PLAN.md — deriving the buy-box standard, building jurisdiction dossiers, swapping the geo stack off Google API keys, or moving screening Phase 3 in-conversation. Encodes what to delegate to which subagent, and the safety gates.
recap
by YashuLanki
Use when the user wants to reorient on this project — where we left off, what's next, what we're trying to do overall — or when they invoke /recap. Good for the start of a new conversation/session.
screening-run
by YashuLanki
Use when asked to screen a CoStar export or broker spreadsheet, rank inbound land listings, decide what to pursue, or when the user invokes /screening-run <file> [moic]
document-research
by YashuLanki
Use when the user wants anything researched out of the firm's documents or the public record — findings from due diligence files, a jurisdiction dossier, or a claim checked before it's trusted. This is the documents desk's lead playbook; it routes to the right worker subagent instead of doing everything in one undifferentiated pass.
proximity-mapping
by YashuLanki
Use when the user wants proximity/POI analysis around a portfolio property or a screened listing — what's nearby, how a candidate compares to what the firm already owns, or a batch of proximity runs across a shortlist. This is the proximity desk's lead playbook; it dispatches worker subagents when a result needs checking, and knows which failure modes to distrust.
commit_git
by YashuLanki
Use when asked to commit and push changes to GitHub, or when the user invokes /commit_git. Reviews what actually changed, stages it deliberately, drafts a commit message in this repo's own style, commits, and pushes to the current branch's tracked remote.
vaulter-screening-pipeline
by YashuLanki
Use when a new CoStar/broker export needs to go from raw file to a trustworthy, shareable result — screen it, verify the screener read it fairly, map proximity on the shortlist, and confirm the report is correct and readable before handing it to a partner. Invoke when screening something "for real" / "to send out", not for a quick exploratory look.
mcp-health-check
by YashuLanki
Use to check whether the Vaulter AI MCP connector itself is running smoothly -- not the data behind it, the server process. Invoke when the user asks to check the connector/server directly, or any vaulter_ai tool call has just errored, timed out, or hung.