CurrybotLC team engineering assistant
Pick the Currybot exit first. Use the lower sections only if the next route is still unclear.
- Primary job: Currybot remains Atlas’s shortest explanation for team-facing code review, engineering Q&A, and delegated technical help across the LC stack.
- Working surfaces: Readers reach Currybot through the registry’s direct surfaces — client-support workflows and the team bridge — instead of a drift-prone single-host runtime story.
- Best use: Open this district when the real question is what Currybot should handle directly, how to ask it well, or when another agent should take the next move.
Reader orientation
Start with the request-shape section
Use the request examples first when you already know Currybot is the right helper and the real problem is framing the ask clearly enough for a strong answer or review.
Best when you need a review prompt, a codebase question, or a delegated technical task phrased tightly enough to avoid vague output.
Stop once you can name the repo, scope, failure mode, or focus area Currybot should use as its starting point.
Currybot is the review-and-support specialist, not the whole engineering lane
Currybot is the review specialist when the next move is checking correctness, risk, or implementation quality rather than writing the feature from scratch.
Currybot answers scoped questions about code paths, architecture, CI failures, and past engineering decisions without requiring the reader to know the whole repo first.
Clawdy and other agents can pull Currybot into the loop when the work needs repo-specific review depth, branch analysis, or a technical second pass.
The registry carries Barq Client Pilot as Currybot’s client-support process charter — paused today, so the digest-and-tracking loop is on hold rather than an active recurring lane.
Cadence process cycle for client-support digest and tracking
Specific asks beat vague prompts
"Look at this PR"
"Review PR #234 in currykit — focus on the auth middleware changes and check for race conditions"
Scope plus focus area gives Currybot the right files, failure mode, and review lens immediately.
"What's wrong with the build?"
"CI is failing on the lint step in lambda-curry/frontend#89 — what's causing it?"
Repo, failing step, and concrete symptom let Currybot work from evidence instead of guesswork.
"Explain the backend"
"How does the Currybox health-check endpoint work? Include the handler path and retry logic."
Specific subsystem plus the exact answer shape keeps the response grounded and fast to verify.
Know when Currybot should hand the next move away
Clawdy owns cross-network coordination, deployment-adjacent orchestration, and the gateway-level conversation that may delegate into Currybot later.
Use Clawdy when the request spans systems, sessions, secrets, or cross-agent routing instead of code review depth alone.
Scout owns bounded implementation cycles, proof-backed delivery, and product-surface improvements that need code written rather than reviewed.
The simplest split: Currybot reviews and supports engineering work; Scout carries the named build slice.
Meg is the design gate for route-level direction, PDLP alignment, and screenshot-audit quality checks that Currybot is not meant to arbitrate.
Use Meg when the problem is UI coherence, design direction, or visual review rather than code correctness.
Atticus owns the ruling layer when the request crosses architecture, policy, approval, or trust-boundary decisions that Currybot should not make alone.
Use this route when the code question has turned into a “should we?” or “are we allowed to?” judgment call.
Continue from Currybot
Currybot keeps this as a street-level exit so the district can stay short and practical.
Currybot keeps this as a street-level exit so the district can stay short and practical.