Run cogs in custom working directory

[juniper-shopify]

Cogs that run external commands, like cmd and agent, or that access files on disk (like chat, when we give it the ability to do thing like handle image input/output) may benefit from being configured with an alternate working directory.

---

Note: This PR does not make Roast automatically change into the configured working directory when running a cog for which it is specified, because doing so causes issues for parallelization. (If two cogs are running in parallel and each attempts to change the working directory of
the main ruby process, then it gets changed for all parallel cogs, leading to unexpected behaviour. There's no way to effectively coördinate this.)

Instead cogs that need to interact with the filesystem, or care about their working directory for any reason, should use the working_directory config value internally to intentionally do whatever they want to do within that folder without changing the working directory of the ruby process.

(We'll maybe want to put some guardrails in place, at least a warning, to prevent the user from inadvertently changing the working directory in, say, a cog's input block. Though that's only an issue if they do use async cogs and don't realize/anticipate the side effects. Probably a niche issue.)

[juniper-shopify]

• Basic streaming messages for Claude agent provider #518
• Expand the agent provider structure and add some configs for the claude provider #517
• Run cogs in custom working directory #516 👈 (View in Graphite)
• Alphabetize function DSL test cases #515
• Add method doc to cogs config #514
• Deep-Dupe config defaults before setting #513
• Add validation to cog configs #512
• Add a global config scope to set basic things that apply to all cogs of all types #511
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[juniper-shopify]

Merge activity

• Nov 3, 4:51 PM UTC: @juniper-shopify merged this pull request with Graphite.