agentlas-ai/agentlas-os/agentlas-packaging

Agentlas Packaging

Procedure

1. Inspect the existing source: prompt, repo, ZIP, runtime folder, skill, command,

or generated agent package.

1. Classify it as single-agent, team-builder, or mixed/unclear.
2. Run contracts/builder-interview-research-gate.md when the source behavior,

target user, tools/plugins, output artifacts, or evaluation bar are unclear. Packaging must not turn a shallow prompt into a polished but weak package. When the gate runs, research official sources, similar agent repositories or comparables, academic/professional theory, and plugin docs before changing the source behavior.

1. Preserve useful behavior while adding Agentlas contracts:

• AGENTS.md;
• docs/builder-interview.md;
• docs/research-sources.md;
• docs/tool-selection.md;
• docs/domain-expert-synthesis.md;
• docs/prompt-performance-contract.md;
• .agentlas/capability-eval-plan.json;
• .agentlas/agent-card.json;
• .agentlas/company-blueprint.json;
• .agentlas/mode-map.json;
• .agentlas/memory-map.json;
• .agentlas/memory-tickets.jsonl;
• .agentlas/vault-references.json;
• .agentlas/global-commands.json;
• runtime adapters;
• verification scripts.

1. Author the workforce résumé block on the routing card (workforce on

.agentlas/agent-card.json / the marketplace routing card). The hub Workforce search matches on exactly these fields; a card without them is invisible to every WorkOrder that uses them, and card lint blocks routing_ready without the block. Use the versioned ontology graph contract (agentlas_cloud/workforce/ontology_v1.json, awo:2026-07-15.2) for seed aliases and relation semantics, never as an allowlist:

• roles: 0-4 open role:* professional-responsibility ids;
• communities: 1-5 open community:* professional-domain ids;
• skills: 3-12 open skill:* verb-object capability ids;
• knowledge: open knowledge:* domain/method ids backed by the package;
• modalities: optional non-text input/output metadata;
• languages: optional genuine delivery languages, never listing locales.

New well-formed semantic IDs are valid graph concepts and must not be rejected merely because the seed snapshot has not seen them before.

1. Add or repair the global command across Claude Code, Codex, Antigravity,

generic AGENTS.md tools, and terminal adapters.

1. Remove secrets, raw logs, private local notes, and unsafe public paths.
2. Run scripts/verify-team-package.sh <package-root> after repair. If it

fails, correct the package shape before any final handoff.

1. Run package verification and public-safety checks before release.

Output

Return classification, repaired_files, agentlas_contracts_added, runtime_adapters, global_commands, verification, and blockers.