COPS repository layout

Path	Responsibility
`README.md`, `docs/GETTING_STARTED.md`	Analyst-first discovery and safe first use
`agents/`	Centralized specialist agent profiles (`profiles/*.agent.md`) and canonical registry (`registry.json`)
`cops/`	Standard-library catalog, validation, command safety, demos, and checks
`cops/contracts/`	Operational contracts runtime, lifecycle state machines, and evidence verification
`cops/scenarios/`	Scenario and provenance registry loaders, queries, and cross-validation
`cops/capabilities/`	Capability reconciliation and truth-in-advertising auditor and matrix generator
`cops/routing/`	Deterministic zero-token intent classifier and Triad execution plan generator
`cops/authorization.py`	Interactive operator authorization gate and cryptographic receipt validation
`cops/evidence/`, `cops/connectors/`	Opt-in evidence contracts and bounded acquisition SDK for repository tooling
`catalog/categories.json`	Stable capability taxonomy
`catalog/plugins.json`	Single canonical package inventory
`catalog/scenarios.json`	Canonical operational cybersecurity scenario registry
`catalog/provenance.json`	Pinned external research and tool provenance registry
`catalog/capabilities.json`	Reconciled four-mode capability truth-in-advertising registry
`catalog/schemas/package.schema.json`	Documented package governance contract
`catalog/schemas/specialist-profile.schema.json`	Specialist agent profile contract
`catalog/schemas/authorization-receipt.schema.json`	Tamper-evident authorization receipt contract
`catalog/schemas/evidence-envelope.schema.json`, `catalog/schemas/acquisition-receipt.schema.json`	Shared evidence and acquisition receipt v1 contracts
`catalog/schemas/-contract.schema.json`, `catalog/schemas/-registry.schema.json`	Operational contracts, scenario, capability, and provenance registry schemas

`examples/evidence-sdk/`	Offline SDK walkthrough and design-only connector request plans
`plugins/<category>/<plugin-id>/`	Self-contained distributed product package
`plugins/logging-telemetry/security-logging-advisor/`	Repository scanning and security logging guidance
`plugins/detection-hunting/soc-investigation-workbench/`	Evidence-backed investigation planning and review
`plugins/detection-hunting/sentinel-hunt-workbench/`	Offline Sentinel hunt authoring, rendering, and stress evaluation
`plugins/detection-hunting/attack-path-workbench/`	Offline, evidence-linked analysis of illustrative attack paths
`plugins/offensive-security/attack-surface-planner/`	Approved-scope, offline attack-surface discovery and passive test planning
`.agents/plugins/`, `.github/plugin/`, `.claude-plugin/`	Generated host marketplace indexes; do not edit manually
`AGENTS.md`, `.agents/skills/`	Canonical contributor contract and maintenance workflows
`.claude/skills/`	Generated contributor-skill mirrors; do not edit manually
`scripts/agent/`, `Makefile`	Contributor doctor, portable export, prerequisite installer, and full repository gate
`specs/`, `.specify/`	Product requirements, acceptance scenarios, and durable context
`tests/step_defs/`	Executable pytest-bdd acceptance steps
`docs/`, `docs/decisions/`	Operator guidance, architecture, governance, and decisions
`licenses/`, `THIRD_PARTY_NOTICES.md`	Project and imported-foundation attribution

Package ownership boundary

A package owns its v1.0.0 root plugin.json, Codex .codex-plugin/plugin.json, Claude .claude-plugin/plugin.json, product skills, optional agents, scripts, examples, schemas, package documentation, and deterministic tests. Shared code belongs at the root only when multiple packages genuinely depend on one stable contract. Cross-package use must name the dependency and fail closed when it is missing or incompatible.

The host-neutral distribution is generated separately from each source package. The export includes approved standard and namespaced files, and omits native host manifests so Claude Code discovery remains available in the source marketplace.

The validator rejects unknown categories, incorrect category-first paths, uncataloged packages, duplicate IDs or skill names, manifest/catalog divergence, unsafe declared commands, unsupported evidence claims, and generated-index drift.

The .agents/skills/plugin-run-cost/ contributor skill provides local retrospective run accounting, input profiles, explicit forecasts and business comparisons. Its synthetic fixtures are public; execution ledgers and customer source exports must remain private outside the checkout. See the skill.