What OpenGantry helps with

Product pitch and problem framing: README. This page maps situations and personas to the GXT workflow — when OpenGantry is the right fit and what you get out of it.

All docs: index.md

When to adopt

You adopt OpenGantry when agent velocity is valuable but unreviewed scope creep, silent governance edits, or unverifiable “it works” claims are unacceptable.

OpenGantry = scoped work order + shell gates + structured verify output
Executor agent = worker (Cursor, Hermes, human dev)

Typical triggers:

vs TDD + GitHub Actions

You already have tests and CI. The friction is not “do we have gates?” — it is who reads the failure output.

Human TDD works because a developer parses stderr and fixes code. When an IDE agent runs the same loop, unstructured terminal output causes hallucinated fixes and infinite retries; someone senior still babysits.

OpenGantry keeps your existing gate_command (e.g. npm test) and adds:

• Mission YAML — declared edit paths before the agent touches files
• findings[] — machine-readable failure envelope (file, line, resolution_hint) from gantry verify --json

Vocabulary and the full contrast table: README § In plain English and README § Why not just TDD and CI?.

Personas

Teams already doing TDD who still babysit IDE agents

Your test suite and CI are solid. Cursor or Copilot still edits outside the intended blast radius, and when a gate fails the model misreads stderr. You need the same gates with structured retry input and Git-locked scope — not another test framework.

See README § Why not just TDD and CI? and FEATURES.md § gantry verify.

Regulated engineering (fintech, health, critical infrastructure)

Auditors ask how AI-assisted coding fits your ISMS or AI management system — not how good your prompts are. OpenGantry produces operational records: Planner [MSN-XXXX] commits, mission scope, executor trace, verify output.

See COMPLIANCE-ISO.md for ISO 27001 / 42001 mapping and ADOPTION.md for the standard change loop.

Platform / platform-security teams

You need blast-radius control before agents get write access: subprocess TMVC via gantry runtime exec, hooks on governance paths, merge-time gantry verify. Enforcement tiers are documented honestly — IDE Write/Edit alone is advisory.

See FEATURES.md § Enforcement boundary and INTEGRATIONS.md.

Autonomous orchestrators (Hermes, CI bots, custom runners)

External executors implement code; OpenGantry owns contract and verdict. On verify failure, ingest structured findings[] — no terminal log scraping.

See AGENT-LOOP.md.

Content and brand compliance

The same mission/verify loop governs TypeScript imports and marketing copy: regex perimeter rules (forbid_pattern, require_pattern) with file + line in the failure envelope.

Walkthrough: examples/content-governance/. Adapter details: DOMAINS.md.

Teams replacing improvised agent scripts

Many teams use IDE agents with no orchestrator; others ship one-off agent-run.mjs files with ad-hoc state and heuristic scope. OpenGantry replaces improvisation with a Git-native protocol envelope — legislative commits, pinned missions, verify-gated workflow.

OpenGantry vs alternatives

OpenGantry is Autonomous Repository Engineering — determinism, predictability, and standardized protocols for scoped agent work. It is not a real-time conversational wrapper and not a cloud observability product (see Gantry.io for that category).

Non-goals: always-on improvisation, unscoped IDE writes as source of truth, replacing your CI — GXT adds a narrow inspectable envelope on top of Git.

Reproducible benchmark

Clone this repository and compare a pedagogical orchestrator specimen against OpenGantry TMVC on the same task:

git clone https://github.com/jeger-ai/opengantry.git
cd opengantry
npm ci && npm run build
npm run examples:benchmark

What this is (and is not): the raw-script phase compresses common anti-patterns into one file. Most teams do not ship a monolithic orchestrator; the benchmark still shows that without a Git-native protocol envelope, structure and auditability are improvised.

Contrast specimens: examples/contrast-agent-script/ vs examples/gantry-minimal/. Full harness: examples/benchmark-agent/.

Machine-readable: npm run examples:benchmark -- --json.

Custom domains without built-in adapters

Accounting, legal, or proprietary corpora: use gate_command scripts and TMVC path globs. Document patterns in your repo’s ARCHITECTURE.md. Built-in adapters cover code and content; everything else follows the same mission loop.

See DOMAINS.md § Recipes without built-in adapters.

Related docs