I've found several solutions to avoid agents from building complexity -- the first is to not use Astra or Fable/Mythos unless absolutely required. The second is to use effort level "high", not "xhigh". The last idea would be to edit your AGENTS.md to be short, and start it with something like:

Contributors to this project must follow YAGNI, DRY, KISS, and principle of least astonishment, both from user-visible surfaces, and backend design.

Code is a liability. Do not build anything not expressly asked for.

1. Minimal, sometimes only a link or a sentry ID and a skill name

2. Both. Context is kept up to date for every nontrivial task in a "technical project plan": https://photostructure.com/coding/claude-code-tpp/

3. If you have separate clients, I would use a different username/login for each client so session transcripts (like in ~/.claude/... are segregated.

4. The more slack I give to agents, the more likely a session's work will have to be discarded. TPP (and looking at them before implementation starts) help avoid these very-off-course sessions from happening at least in my experience.

5. I've been iterating on coding and reviewing skills, and also just released the local reviewboard/PR tooling I've been using for the past couple of weeks. Everything is open source: fork and tweak to your content.

https://photostructure.com/coding/

I don't buy it. You can steer them gently to at least write human-accessible project plans—and then session compaction, agent switching, and review is much more doable. I've been iterating on this for over a year: https://photostructure.com/coding/claude-code-tpp/

I switched to the Linux desktop as soon as it was available to avoid all of the TUI rendering bugs and restore sensible copy/paste. I was using vscode's git panel to review diffs before they got committed, but had a feeling that there had to be a way to make agents pull harder so my review wasn't always the long pole in the tent.

Imagine a local GitHub PR web UI (thanks to difit), married to your agent for real time in-context discussions, and built-in skills to teach agents how to write easy to follow PR, focusing on both why's and hows.

Last night I released gavel that scratches this itch: it's open source, so fork and enjoy. https://photostructure.com/coding/gavel/

The author briefly mentions that nits break the reader's flow state. Your comjicaito only proves the road navagable—not pleasant.

Can't you do a .well-known or DNS handshake (like certbot) so people prove domain ownership, then hand then the report of what you found on their domain?

*0

Have your agents keep contexts of tasks up to date as they work on stuff.

I've been iterating on this approach for about a year—it makes switching between agents and harnesses trivial.

https://photostructure.com/coding/claude-code-tpp/

"use subagents as applicable" (and maybe "use sonnet for simpler tasks")

Thanks. The Sonnet part is smart to save tokens.

Same on the other side of the pond