example.agent-application-foundation

@lssm-tech/example.agent-application-foundation

Filesystem-authored reference app for the agent application foundation lane.

What it demonstrates

• Compiling agent/{instructions.md,agent.ts,tools,skills,connections,sandbox,channels,subagents,schedules} plus evals/ with @lssm-tech/agentpacks/application.
• Running a credential-free orchestrator-plus-verifier eval against a single-agent baseline with @lssm-tech/lib.agent-evals/runner.
• Executing the CommunicationOS-shaped draft → approval wait → send → receipt lifecycle, including an exactly-once replay proof for the side effect.
• Exercising the local development sandbox honestly: network allowlist claims are rejected and durable artifacts are extracted and reloaded without making an isolation claim.
• Composing the EVE adapter with the independently conformance-tested production DockerAgentSandbox, while keeping the native EVE sandbox's narrower losses explicit.
• Exercising both public adapter targets: Workflow DevKit plus EVE, with replay, cost, latency, and redaction evidence without serializing secrets.
• Composing credential-free enterprise topology choices for Vercel + Supabase, Railway + Supabase, Vercel + Railway + Supabase, and Docker isolation while preserving provider roles, unresolved operator requirements, and candidate-only qualification truth.

Enterprise topology example

Import composeReferenceAgentPlatform from @lssm-tech/example.agent-application-foundation/enterprise-platform. The returned plan includes explicit enterprise controls and a qualification report; it remains a candidate until the consumer supplies and passes live suites for the exact environment.

All-in-one agents

The package includes two credential-free product scenarios in addition to the low-level reference fixture.

Autonomous Release Steward

Mission: keep a consumer codebase compatible with external APIs and dependencies by producing safe, verified, human-approved migration pull requests.

bun run example:release-steward

Import runReleaseStewardExample and RELEASE_STEWARD_STAGE_IDS from @lssm-tech/example.agent-application-foundation/release-steward. The result includes the immutable source digest, affected usages, isolated subagent evidence, digest-bound approval, exactly-once patch receipt, CI and draft-PR evidence, durable worker and governed-improvement evidence, deployment composition, and candidateReadiness. It never auto-merges.

The filesystem source is also a Protocol V1 application with the stable installation ID release-steward. signReleaseStewardApplicationV1 accepts an injected signer so private key material stays in the caller's KMS/HSM or signing service. Local examples remain unsigned unless the caller explicitly composes that port; a test key never qualifies a production installation.

Company Intelligence Scout

Mission: continuously build source-cited company, offering, mission, goal, location, job, professional-person, and eligible public-contact intelligence.

bun run example:company-intelligence-scout

Import runCompanyIntelligenceScoutExample from @lssm-tech/example.agent-application-foundation/company-intelligence-scout. Fixture mode reads only committed content and performs no network request or outreach. The default autonomy mode is draft-only; retained person/contact records expire after at most 30 days. Live discovery, fetching, rendering, suppression, and outreach require explicit policy and consumer-supplied ports:

bun run example:company-intelligence-scout:live -- \
  --policy ./company-scout.policy.json \
  --ports-module ./company-scout.ports.ts \
  --seed-url https://allowed-company.example

Publicly visible personal data is not unrestricted data. Consumers own purpose, lawful basis, transparency, rights, suppression, deletion, jurisdictional requirements, and live provider qualification. Automated hiring decisions, candidate rejection, job applications, protected-trait inference, and sensitive-data enrichment stay outside the example.

Start with the public Agent Ecosystem guide, then use the repository enterprise agent platform guide^† for the complete copy-ready source. Both cover multiplayer, governed improvement, continuous loops, self-maintaining APIs, provider composition, and production qualification.

Local commands

• bun run build
• bun run typecheck
• bun run test
• bun run preflight
• bun run example:release-steward
• bun run example:company-intelligence-scout