Portable agent expertise
Your AI coding agent, grounded in Mappls.
Install one standards-compatible skill that routes planning, implementation, debugging, migration, stateful journeys, MCP, A2A, and release handoffs through the same source-backed catalog as this developer platform.
Verify before install
Portable files. Inspectable instructions. No bootstrap magic.
The ZIP contains a single build-with-mappls folder. Verify its published digest, inspect the Markdown and zero-dependency script, then place the folder in the skills directory supported by your agent client. The open format requires SKILL.md; client-specific discovery locations vary.
This package follows the current required Agent Skills frontmatter and folder rules. It does not claim experimental registry, signature, OCI, or automatic-install proposals as standardized behavior.
curl -fsSLO https://developer.mappls.com/downloads/build-with-mappls-skill.zip
curl -fsSLO https://developer.mappls.com/downloads/build-with-mappls-skill.zip.sha256
shasum -a 256 -c build-with-mappls-skill.zip.sha256
unzip build-with-mappls-skill.zip
# Copy build-with-mappls into your agent client's skills directory.
# Keep the folder name unchanged so it matches SKILL.md.Five evidence gates
From intent to a reviewable handoff.
The skill gives the model freedom to solve the application problem while fixing the fragile parts: source selection, credential placement, durable state, evidence labels, and production authority.
Select exact capability
Resolve industry, platform, product, state depth, source generation, and evidence gaps before writing code.
Freeze an integration manifest
Generate one canonical use-case × platform × operating-mode contract before implementation.
Implement the narrowest path
Use exact quickstarts, contracts, SDK records, source launchpads, tutorials, and maintained apps without inventing provider syntax.
Own durable state explicitly
Model identities, transitions, idempotency, concurrency, reconciliation, audit, privacy, and recovery outside model context.
Qualify and hand off
Prove fixtures first, then separately validate issued-account behavior and independent production authority.
Progressive disclosure
Load only what the task needs.
Agents discover a short name and description, read the core workflow only when triggered, then open focused references or query the bundled catalog on demand.
SKILL.mdTrigger metadata and the six-stage source-grounded build workflow.
agents/openai.yamlHuman-facing skill name, description, and safe default prompt.
scripts/mappls-context.mjsZero-network summary, search, and exact-manifest helper.
references/catalog.json714 destinations, 624 manifests, and 13 platform starters.
references/authority.mdCredential, source, entitlement, compatibility, and release truth boundary.
references/stateful.mdIdentity, transitions, replay, concurrency, reconciliation, audit, and recovery.
references/ai-native.mdMCP, A2A, retrieval, model-visible data, evaluation, and human control.
Real build prompts
One skill, very different systems.
“Add address autocomplete, serviceability, and an explainable delivery promise to our React storefront.”
“Build an Android technician journey with assignment, route preview, offline proof, independent review, and replay-safe sync.”
“Design a read-only copilot that inspects a delayed vehicle and prepares a human-approved recovery route.”
“Plan an offline navigation package update with A/B slots, power-loss recovery, qualification, and rollback.”
“Move our legacy iOS maps integration to the exact current source without inventing package compatibility.”
“Create an evidence-led retail site analysis flow with immutable datasets, scoring lineage, sharing, and revocation.”
Negative authority
Useful without pretending to be production.
The skill packages repository-backed planning and implementation guidance. It cannot prove entitlement, live provider behavior, compatibility, support, customer authority, or production approval.