Skip to content

refactor(agents): clarify existing-contract autonomy and simplify role prompts - #1

Merged
abocha merged 1 commit into
mainfrom
fix/role-autonomy-and-prompt-clarity
Oct 8, 2026
Merged

abocha merged 1 commit into
mainfrom
fix/role-autonomy-and-prompt-clarity

Conversation

@abocha

@abocha abocha commented Oct 8, 2026

Copy link
Copy Markdown
Owner

Why

GPT-6-class agents may interpret mentions of concurrency, persistence, security or contracts as automatic stop triggers, even for routine fixes that preserve approved invariants. The five role prompts also repeat generic workflow prose that the injected AGENTS.md already governs.

Changes

  • Permit bounded and integration implementers to complete corrections within established contracts, including high-risk modules; escalate only when a material design/invariant/ownership/authorization decision is actually needed.
  • Shorten all five role prompts while preserving role boundaries, risk-sensitive review, focused validation, leaf-worker restrictions and explicit evidence reporting.
  • Keep the exact model assignments, efforts (Luna xhigh, Sol medium/high), sandboxes, and portable workflow modes unchanged.

Validation

  • Compared rewritten prompts against the original role contracts; all five TOML headers, model/effort and sandbox values preserved.
  • Only five .codex/agents/*.toml files changed. No code, installation scripts or workflow policy modified.
  • Hosted scripts-ci result to be checked before merge.

No Linear policy changes or new process steps.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant