Skip to content

Add decision gate constraints and review checklist#5

Open
khalidelborai wants to merge 3 commits into
me2resh:mainfrom
khalidelborai:feature/decision-gate-constraints
Open

Add decision gate constraints and review checklist#5
khalidelborai wants to merge 3 commits into
me2resh:mainfrom
khalidelborai:feature/decision-gate-constraints

Conversation

@khalidelborai
Copy link
Copy Markdown

Summary

  • Added cognitive constraints to all tool integrations that enforce better
    decision quality before creating an AgDR
  • Created a review checklist for evaluating AgDRs during pull requests
  • Updated README with a link to the review checklist

Why

I noticed agents often rush through decisions — picking the first option,
listing vague pros/cons like "faster" without numbers, or skipping tradeoffs
entirely. These five constraints act as a gate that must be resolved before
the AgDR workflow proceeds.

What changed

Added a "Decision Gate Constraints" section to all 8 tool configs
(Claude Code, Codex, Cursor, Copilot, Windsurf, system prompts) with:

  1. Understand the real problem first
  2. Compare at least 2 viable alternatives
  3. Back claims with evidence
  4. Use measurable specifics, not vague language
  5. Name the tradeoff you're accepting

Also added agdr-review.md with a checklist and three adversarial
questions for reviewing AgDRs in PRs.

Test plan

  • Constraints render correctly in each tool config
  • Review checklist renders on GitHub with working checkboxes
  • README table link to agdr-review.md works

Add 5 inhibition-based constraints to the decision workflow:
- Do not document before understanding the real problem
- Do not accept the first option — compare 2+ viable alternatives
- Do not list unverified pros/cons — base claims on evidence
- Do not write "better"/"faster" without measurable criteria
- Do not proceed without naming the accepted tradeoff

Applied to: commands/decide.md, skills/decide/SKILL.md,
tools/codex, tools/cursor, tools/copilot, tools/windsurf,
tools/system-prompts
Inhibition-based checklist for reviewing AgDRs in pull requests
with 3 adversarial review questions.
Copy link
Copy Markdown
Collaborator

@atlas-apex atlas-apex left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks so much for this — really appreciate the thought you put in, and it's clear you read the existing spec carefully before writing. The five decision-gate constraints are a meaningful hardening of the softer "Tips for Writing Good AgDRs" guidance in the template, and the agdr-review.md checklist fills a genuine gap I hadn't filled yet. The three adversarial questions in particular ("what would make this decision wrong?") are excellent — that inversion catches the kind of pro/con blindspot the current template makes too easy.

I'd love to merge this. Two small things before I do, and one note I'll leave for later:

Please add:

  1. CHANGELOG.md entry under an "Unreleased" heading noting the addition of the decision-gate constraints + review checklist. The repo maintains a CHANGELOG by hand and this is a user-facing change.

  2. One sentence about agdr-template-short.md — ideally in the constraints themselves, or as a clarifying note alongside them. The "≥2 alternatives must be compared" rule is a hardening of what used to be a soft suggestion, so any existing AgDR that compared only one option is now retroactively non-compliant. That's fine for the long template, but the short-template is explicitly designed for simpler decisions; worth saying whether the ≥2 rule applies there or not. Either answer is defensible — I just don't want the ambiguity.

Note for a future PR (not blocking):

The five constraints are duplicated verbatim across 8 integration files (Codex, Cursor, Copilot, Windsurf, system-prompts, commands/decide, skills/decide, agdr-review itself). That matches the existing pattern in the repo where each integration gets its own copy of the guidance, so I'm not asking you to refactor it here — but the next time we update these constraints we'll need an 8-file sweep. If you want to tackle that extraction in a separate PR I'd be keen to see it.

Otherwise, the PR is in great shape: scope is tight, voice matches the rest of the spec, commit history is clean, and the description has all the sections CONTRIBUTING.md expects. Once the two asks above are in, I'll merge. Thanks again for contributing.

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.

2 participants