Sherpa-B Reference

transform your LLM call inventory into a clean agent architecture with clear boundaries, control loops, and handoff contracts

compare your codebase against its specifications to identify gaps and drift, then create GitHub issues for everything that needs attention

select the right issue to work on, create a branch, write tests first, and implement the solution

decide whether you should raise at all, then walk into any investor conversation with a default alive position, an evidence-based traction narrative, a specific ask, and a walk-away number

process notes from a demo, user-testing session, mentor chat, or social media reaction - talk through what they actually mean for the project, then log them into the shared feedback checklist coach and other skills read

Long-lived work coordinator - surface Sherpa-B tasks and this repo's GitHub issues together, help decide what's next, hand off the actual work to a second terminal. The goal is to help builders stay on top of tasks and deliverables.

design how information flows through your LLM calls, ensuring each reasoning step gets exactly what it needs while avoiding context bloat and quality risks

close out a session outside any workout - commit changes, reflect, persist notable decisions/facts/preferences, and recap what got logged

capture, structure, and create GitHub issues from conversations, code reviews, or feature ideas so nothing gets lost

turn a workout reflection, planning doc, or general observation into a ready-to-publish social post in the participant's own voice

build a compelling 8-minute demo narrative that resonates with your target audience - craft your story arc, storyboard your visuals, develop your script, and prepare for recording

transform your agent architecture into a diagnostic evaluation system by validating context inputs structurally, evaluating outputs semantically, and tracing failures back to context management issues

design systematic evaluation for your AI agent through risk-driven test case creation

finalize your PR with spec updates, follow-up issues, and code cleanup, then mark it ready for review

deploy your agent to a small group of testers to validate your quality risk assumptions through real-world usage

frame the change I want to make, explore how to build it, and plan how to validate it

periodically audit what you have actually built - where it ended up living, and what it is built on - and leave with a ranked list of what to keep, simplify, consolidate, or pay for

Get a short-lived link to open the Sherpa-B web dashboard in a browser, optionally to a specific page

First-time onboarding for bootcamp participants - sets up workspace and gets to know them

Persist notable info from this session to telemetry/prose docs, ad hoc - not tied to a workout checkpoint

Bind this repo to a Sherpa-B project by running init with a real git identity

review your own code changes with quality checks before requesting peer review

Check that orientation's plugin setup (telemetry install, seeds, settings) is actually in place, and fix anything missing

Roleplay a skeptical persona (customer, CTO, investor, employer, or others) grilling you on your project. Fetches the persona live from the Sherpa-B MCP server every run.

rehearse defending your codebase against a roleplayed skeptical persona (customer, CTO, investor, employer, or others) before a presentation, surfacing blind spots while it's still safe to find them

prepare to record a 4-5 minute sprint demo video - verify the app works, draft a timed script, sketch diagrams, and get recording-ready

submit a recorded demo video for review - confirm the share link is viewable by anyone, gather app name/description/feedback ask, and submit via MCP

design your agent's user experience by mapping how user context flows into agent context and creating implementation-ready UX specifications

find and talk to real users, learning directly about their needs, current solutions, and biggest pain points

lock down what you are, for whom, and against what alternative - a positioning statement and one-page brief that governs every downstream GTM decision

read whether your business has product-market fit by scoring five hard signals - retention shape, the 40% test, organic vs. pushed growth, revenue quality, and default alive - then write the pivot trigger before emotion enters the room

establish your real financial position - runway, unit economics, deal size floor, cash flow architecture, and funding path - before making any GTM decision

turn a vague sense of "who our customer is" into a falsifiable first-customer hypothesis - job-to-be-done, cost and frequency of the problem, buyer authority, trigger - specific enough to name 10 real people and test it in three conversations

manually acquire your first customers before automating anything - map your existing access, rank targets by pain x reachability, write ten personalized messages, run the sprint, and extract the patterns that decide whether you are ready to systematize

turn the manual outreach signal from your unscalable acquisition sprint into a repeatable engine - ICP signal card, message architecture, cadence map, volume model, and a minimal tooling stack you can operate alone

prepare for a specific upcoming customer conversation, drill Mom Test question discipline, and convert the call into a dollar-quantified pain statement, gap map, and qualification verdict

turn sales pushback into a classified objection map, non-collapsing response scripts, and a closing sequence that ends in a named commitment

set a first price you can defend with logic - anchored to discovered customer value, structured before it is numbered, and designed as a market experiment with pass/fail signals rather than a permanent commitment

design the pilot before it starts - conversion path, customer-owned success metrics, a defensible opt-out clause, cash-flow-safe payment terms, and an expansion motion wired into the agreement

start a session by reviewing recently logged telemetry - kind/domain/risk_dimension/actor-role stats - and let the participant correct anything stale or wrong before diving in