OrangePro boundary

Type: agentic-system-analyses/types/agentic-system-boundary.md

Boundary and evidence

The target is the shipped OrangePro repository: its local evidence graph, test-drafting and proof workflow, CLI and MCP interfaces, agent-facing instructions, platform API adapters and ancillary deployment code. Its intended use is to identify test gaps in a checkout, supply grounded test drafts and distinguish test associations from targeted mutation evidence. This is an extension or tool mechanism, not the enclosing coding-agent runtime. Its model-dependent generation and candidate-link lanes qualify it for analysis even though analysis, ranking and proof can operate without a model. SRC-1: package.json, src/local/operations.ts, src/local/mcp.ts.

The boundary is whole-system for this shipped artifact, not for OrangePro's remotely hosted platform or the user's coding environment. The package binaries enter dist/local/cli.js; the launcher resolves that same engine. The repository also contains a separate platform stdio entry and a remote HTTP worker. These shipped paths remain included rather than being mistaken for external applications. SRC-1: package.json, packages/mcp-server/bin/mcp-server.js, src/index.ts, src/server.ts, packages/mcp-remote/src/index.ts.

const cliPath = require.resolve("@orangepro/orangepro-mcp/dist/local/cli.js"); --- packages/mcp-server/bin/mcp-server.js

Initial tracing covers production, retention, admission and later consumption of local state. opAnalyze builds a graph from source and enrichers, preserves generated artifacts and candidate flows, and saves it. opGenerate reloads that graph, resolves a provider, invokes generation, persists fresh drafts and refreshes reports. The start driver invokes analysis, optional AI linking and flow discovery, automatic proof and gap drafting. Proof results enter a ledger, and traceability/report consumers reload graph and ledger. These are shipped responsibilities, including automatic test execution, not solely responsibilities of an external coding agent. SRC-1: src/local/operations.ts, src/local/workspace.ts.

const closed = dynamicProofSucceeded(oracle); --- src/local/operations.ts

const result = buildRtm(graph, loadLedger(root), { targetIds, changedFiles, statuses: opts.statuses, limit: opts.limit, scope }); --- src/local/operations.ts

The agent workflow also instructs a host agent to write and run returned drafts and submit proof arguments. This host-driven path coexists with the shipped automatic and assisted paths; it does not justify classifying the repository as only a returning library API. The skill's broad non-execution wording is documentation evidence, not a boundary that removes the executable proof drivers. SRC-1: src/local/agentWorkflow.ts, rules/orangepro.mdc, skills/opro/SKILL.md, src/local/operations.ts, src/local/draftTest.ts.

"Write each runnable test body to its suggested_path, run its run_command from the owning package, then call the returned prove_run args after a pass to attempt public Proven. Use record_run only for static diagnostics.", --- src/local/agentWorkflow.ts

The revision is frozen at 039cd936e38eb7d52d0e03ded3b3168373d55d77, verified as the checkout's HEAD. The applicability cutoff is 2026-11-08, the supplied run date; it makes no claim about later revisions. The overall evidence tier is code-grounded. Inspection is static: no target execution, provider call, test run or service deployment was performed. Test source is evidence of encoded checks, not observed passing results. Documentation and comments supply declared contracts, not independent demonstrations.

External dependencies include the coding-agent/MCP host, Node and package dependencies, customer checkout and its runners, configured model endpoints, Docker for the containerized verification lane, GitHub for delivery, and the OrangePro backend for platform tools. Host planning, host context management and user authorization are not attributed to this target. The backend implementation is not in the frozen source: adapter inspection cannot establish backend scheduling, model prompts, knowledge-graph persistence or tenant isolation. Provider service internals and customer runs are likewise unavailable, preventing conclusions about actual model quality, live reliability, cost and end-to-end effectiveness. SRC-1: package.json, src/local/generate/providers.ts, src/local/actionsRun.ts, src/local/publishTests.ts, packages/mcp-remote/src/index.ts.

Shipped-path coverage

Included means within the functional analysis boundary, not exhaustively read. The initial inspection follows the material entry points and state loop above. Remaining helpers, release prose and fixtures are available at this same pin for later inspection; they are not access gaps. All rows refer to SRC-1.

Source register