FAQ
Can I use this with an existing project?
Yes. Install the plugin as described under Installation and start a session in the project. The framework is additive: it adds ops/ and doesn't modify your existing code.
Do I need all the CLIs?
The core trio (Claude, Antigravity, Codex) is the supported baseline; run at-setup to get them live. From the optional tier (OpenCode, Kimi, Cursor, Devin, Grok Build) you enroll any subset through at-setup, and an absent one is skipped cleanly in every roster fallback chain. Claude alone can still run the pipeline in a degraded mode, without the multi-model review, test and analysis lanes.
Do I need all the skills?
No. Skills activate contextually. If you never cut scope, the scope-cutting skill won't activate. You can delete any skill directory you don't want.
How do personas differ from skills?
Skills are methodology documents that guide an agent's behavior. Personas are separate processes that dispatch_persona starts, each with its own context and only the tools its manifest entry allows. Skills can be injected into any agent (including external ones like Antigravity and Codex).
What are Agent Teams?
Agent Teams spawn multiple Claude Code instances that collaborate through a shared task list and messaging. Review swarms only analyze; Agent Teams are peers that divide file ownership and coordinate builds. Enable them with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: "1" in settings.json.
Do small bug fixes need the full pipeline?
No. Use at-quick for changes touching fewer than 3 files. It skips Phase 0, plan validation, and the full review swarm.
Can a sprint run with nobody watching?
Yes. at-ship runs every phase in one session, and scripts/coordinate.sh runs the sprint from a terminal across fresh sessions until ops/.sprint-complete exists or the session limit is reached. A Codex lead needs your --allow-full-access on the command line. See Unattended runs.
How does context exhaustion recovery work?
Two layers. Inside a session, the ops/.sprint-complete sentinel is the authoritative completion signal, created only after the verification checklist passes. Under a Claude Code lead, coordinate.sh leads every composed prompt with a /goal line, and at-ship/at-coordinate print a copyable one for interactive runs, but /goal is an assist (CC-03 went 0 of 3 on 2.1.318) and ops/.sprint-complete plus coordinate.sh decide completion. Outside a session, coordinate.sh spawns fresh lead processes with clean context windows and detects completion through the sentinel; state persists via git and ops/STATE.md. A PreCompact hook checkpoints STATE.md before context compaction.
What is knowledge compounding?
After solving a non-trivial problem, at-compound saves it as a structured document in ops/solutions/. Later at-plan and at-deep-research runs search this directory before they start new work.
How do I update Triforge?
Refresh the marketplace, then update the plugin: claude plugin marketplace update agent-triforge and claude plugin update agent-triforge under Claude Code, or codex plugin marketplace upgrade agent-triforge and codex plugin add agent-triforge@agent-triforge under Codex. Any change to the plugin's contents, this update included, asks for approval again, and until then a Codex lead runs without the plugin's hooks. See Update.
How do I move a project from 3.x?
Follow Upgrading from 3.x: the slash commands became at- workflows (command map), the project's instruction file is now its root AGENTS.md, and the agents became personas. Rolling back to 3.3.3 covers the way back.