Skill
Agentic design pattern selector
decision matrix mapping task/context type to the right combination of patterns from the 20-pattern agentic design taxonomy; most real systems combine 3-5 patterns; includes per-pattern TL;DR, cons, canonical combos, and dangerous-combination guard
Primitives inside (10)
compress-tool-outputs-before-contextdisciplineCompress or summarize tool outputs before injecting them into the LLM context - raw tool payloads balloon token count fast.
When: An agent loop feeds tool results back into the model context across multiple steps.
critic-rubric-validation-before-loopdisciplineTest a critic's rubric on known-good and known-bad samples before trusting a reflection loop - a bad rubric produces confident-but-wrong revisions.
When: Setting up an automated critic whose structured feedback drives revisions.
deterministic-routing-confidence-fallbackdisciplineRoute requests with a deterministic confidence score, not an LLM self-rating; below threshold, ask a clarifying question or send to a fallback handler instead of routing.
When: An intent router dispatches incoming requests to specialist agents or pipelines.
fallback-ladder-priority-ordercalibrationOrder recovery fallbacks cheapest-first: simpler method, then cached/saved data, then a safe default answer, then human-in-the-loop.
When: Designing the permanent-error recovery path of an agent or pipeline.
feedback-learning-requires-evaluationfailsafeNever let an agent update its prompts/examples from feedback without an evaluation loop - it can learn the wrong thing from noisy or malicious signals.
When: An agent adapts its behavior automatically from user feedback, reviews, or outcome signals.
interagent-comms-production-avoidancecalibrationFree-form inter-agent messaging (beyond hub-and-spoke) has no known good production record - avoid unless you have enterprise-grade ops capacity; a shared bulletin board is the most tractable topology.
When: Considering direct agent-to-agent messaging protocols for a multi-agent system.
plan-constraint-check-upfrontdisciplineCheck hard constraints (data availability, authorization, budget, deadlines) BEFORE executing a plan - a late-invalidated plan is wasted work and mid-execution replanning is expensive.
When: An agent decomposes a goal into a dependency-ordered plan about to be executed.
pre-prompted-buttons-over-open-textboxgotcha-fixFor customer-facing LLM apps at scale, prefer canned prompt buttons over an open text box - kills the injection vector entirely and tightens context-engineering control.
When: Shipping an LLM feature to thousands of external/untrusted users.
prompt-chain-step-capcalibrationCap sequential prompt chains at 3-5 steps and log per-step artifacts - beyond that, context explosion and error propagation dominate.
When: Designing a multi-step sequential LLM pipeline (ETL, content generation, data transformation).
reflection-loop-max-retry-capfailsafeEvery critic-revise (reflection) loop gets an explicit max_retry cap plus a quality bar that ends it early - never loop until 'good'.
When: A generate-critique-revise loop drives quality improvement automatically.
Get the whole skill
All 10 primitives of this skill as one package, with the order to apply them.
Buy only the primitives you need
Each primitive is 1 credit (≈ €0.10). Pick them from the list above — the button is next to each one.
Upgrade your own skill
Paste your skill; we pick the 5 primitives from the shelf that fit it best, as one bundle for 5 credits (≈ €0.50).
Upgrade my skillNeighbour skills
Skills whose primitives are closest to this one (bge-m3 similarity):