What are you building with Citadel?

[SethGammon]

Hey all, I just enabled discussions. If you're seeing this, I want to hear about what you're up too.

Whether you're just getting started or already running real campaigns with Citadel, introduce yourself here.

A few useful things to include:

• what you're using Citadel for
• what model or provider you're running
• what kind of workflow or codebase you're applying it to
• what is working well
• what feels rough, confusing, or missing

I read every reply. This thread will directly shape what gets improved next, what gets documented better, and what gets built.

I'll start.

Citadel began as a personal tool for coordinating multi-session refactors on a 180K line codebase. The core problem was simple: each new session started half-blind. Agents lost context, repeated work, and stepped on each other. The harness solved that. It gave sessions continuity, structure, and a way to build on prior work instead of resetting.

Now it runs on Claude Code and Codex. The next step is pushing it further for real team workflows, broader research and engineering campaigns, and stronger coordination across agents.

What are you using it for?

[Acquiredl]

Hey Seth, I've been using citadel for two projects now with totally different codebases. I asked claude code to give me answers on my workflow with citadel and here is what I got. I added some notes to some answers. My workflow/citadel usage might be flawed so if some of my assessments are user error feel free to let me know.

What's working well:

1. Campaign state persistence is the killer feature. I've run campaigns spanning 7+ phases (Phase 3 through Phase 7 on Delted Guilds) where each phase builds on the last. The .planning/campaigns/ files keep context alive across sessions in a way that raw conversations can't. Phase 6 (Technical Excellence) completed 7 phases in a single day — refactored a 2100-line combat engine, added error boundaries, split monolithic type/constant files, added worker error handling — all tracked as a campaign with end conditions per phase.

2. /do as the universal entry point removes the cognitive load of "which skill do I invoke." I just say what I want and it routes. Most of my usage goes through /do.

3. /session-handoff is critical for my workflow. As a solo dev working across multiple sessions per day, the structured handoff docs (I have 3 from a single day) let me pick up exactly where I left off.

4. Hooks are genuinely useful. I have 7 active hooks (protect-files, post-edit, pre-compact, quality-gate, circuit-breaker, restore-compact, intake-scanner). The quality gate and circuit breaker catch problems before they compound. The protect-files hook keeps the agent from accidentally modifying its own configuration.

5. The harness.json configuration is clean and practical. Dependency guardrails (banning React Context in a Zustand project) have prevented real mistakes.

6. 25 skills installed and the registry check in /do (Step 0) makes adding new ones seamless.

What feels rough or could be better:

1. Campaign files can get stale. I have a Phase 7 campaign file where Phase 1 status still says "design-complete" when it's actually complete. There's no automatic status sync — you have to remember to update it, or the next session reads outdated state.

2. Cross-project portability. I installed the same 25 skills on both projects manually. Some kind of "Citadel init from template" or shared skill registry would save setup time for new projects. reviewed quickstart.md and use the persistent recommended method using plugin marketplace

3. Windows path handling occasionally trips things up. Not a showstopper but it's noticeable on Git Bash / VS Code terminal.

4. Visibility into what hooks actually do during a session could be better. They run silently — sometimes I'm not sure if a hook fired or was skipped. A debug/verbose mode for hooks would help.

5. Fleet / parallel agents — I haven't pushed this hard yet. The concept is appealing but the use case hasn't naturally come up for my solo workflow.

[SethGammon]

Hey @Acquiredl super appreciate you taking the time to both use and reply!

This is incredible feedback, and my next PR will be based off of it.

Campaign state persistence is the one I'm most proud of and you're pushing it harder than almost anyone I've seen. 7 phases on Delted Guilds is serious. The fact that Phase 6 refactored a 2100-line combat engine in a single day across 7 phases validates the architecture. This is great!

On the rough edges:

1. Campaign staleness is real. Automatic status sync is on my radar. Right now the tradeoff was simplicity (flat files you can read and edit) vs. automation, but you're right that at 7+ phases it breaks down. (I will address this)

2. Glad you found the plugin marketplace for cross-project portability. That's exactly what it's for. If the install flow felt clunky or the docs were unclear, I'd love to hear that too.

3. Windows path handling: I've hit this too. Not 100% sure on the root cause yet, could be a hook or gate resolving paths differently. If you notice a pattern (like which hook or which step trips it up), flag it and I'll dig in.

4. Hook debug/verbose mode is a great call. Silent hooks are a feature until they're a mystery. I'll look into a flag that logs hook activity during a session.

5. Fleet will click when you hit a task that decomposes naturally into parallel work. Think: "refactor 6 modules that don't depend on each other." When that moment comes, you'll know.

---

Seriously, this is great feedback and much appreciated. I'd love to hear more about your project, your work, and how Citadel is helping (or where to improve). Keep pushing it!