architecture-proposal-review
Skill0 starsChanged 58 days agoMITView on GitHub
Review an architecture proposal against repository evidence, integration surfaces, alternatives, and engineering principles before implementation. Use when a user asks to critique a proposed design or architecture. The review is conversational and read-only unless the user requests an artifact.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- implement-with-solSkill
- implement-with-opusSkill
- new-featureSkill
- pre-implementation-analysisSkill
- root-cause-analysisSkill
- worktree-workflowSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
Use heddle/resources/decision-routing.md for ownership and question quality.
- Distinctive: line 21
Use the proposal documents named by the user.
- Distinctive: line 35
Treat every material factual claim as untrusted until checked.
- Distinctive: line 80
Place the options on a benefit-versus-effort frontier.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Architecture Proposal Review
Challenge a proposed architecture before implementation. Determine whether the proposal is factually grounded, serves its real consumers, uses the right system boundary, and offers a sound delivery shape.
Decision ownership
Use heddle/resources/decision-routing.md for ownership and question quality;
feature kickoff delivers the resolved host policy in full. Preserve this skill's
analysis/authoring boundary and applicable standing authorization. A review or
diagnosis request alone grants no implementation authority.
Inputs and boundary
Use the proposal documents named by the user. If no identifiable proposal is available, ask for it; otherwise derive the change scope from the proposal and repository. Read the affected code, tests, governing records, and current integration points rather than accepting the proposal's account of them.
The review is conversational and read-only by default. Write or modify an artifact only when the user explicitly requests one. Use the destination they provide, or ask for one. Reviewing a proposal never grants authority to implement it or to perform external actions.
Review method
Establish reality
- Treat every material factual claim as untrusted until checked. Record it as
verified, contradicted, or unresolved, with repository evidence
such as
file:line, commands, or observed interfaces. - Trace upstream dependencies, downstream consumers, and the blast radius across production code, tests, fixtures, configuration, stored data, and in-flight migration or compatibility work where relevant.
- Extract the concrete requirements and invariants of actual consumers, especially the first or blocked consumer. Do not substitute the proposal's summary for their governing contracts.
Check the system boundary
Sweep the integration surface for adjacent or overlapping mechanisms, including parallel implementations and read/write twins. Give every relevant mechanism one disposition:
- absorb — the proposal should incorporate it;
- exclude — it is intentionally outside the boundary, with the reason stated; or
- missed — the proposal failed to account for it.
Recommend unification only when real callers demonstrate a shared shape. Coincidental similarity is not enough.
Assess the proposal against docs/workflow/engineering-principles.md. Cite
decisive principles by name and link the observed facts to the judgment; do not
copy a second standing-principles register into the review.
Apply "One owner per authority" to competing policy, state or write owners; "Dependencies point inward, not back out" to actual boundary crossings; and "Extend through data and one resolver, not parallel paths" to proposed extension seams. Cite the governing definition and observed consumers, not the label alone.
Compare credible choices
Construct materially different alternatives instead of polishing variants of one idea. Include the smallest intervention that meets the robustness floor and, where useful, a staged path to the same end state. For each option, compare:
- mechanism and ownership boundary;
- fit to verified consumer requirements;
- benefit, effort, and blast radius;
- important principle conflicts and failure risks; and
- reversibility of both the design and its rollout.
Place the options on a benefit-versus-effort frontier. Say plainly when the proposal is dominated by a lower-effort or higher-benefit option.
Separate architecture from delivery
Give an architecture verdict and a delivery verdict independently. A sound target architecture can still have an unsafe migration, sequencing, or cutover plan; a convenient delivery plan does not rescue a weak architecture.
Surface owner decisions separately. For each, explain the options, practical implications, and recommendation. Do not silently overrule ratified product or scope choices. Put defects that exist regardless of the proposal in a separate discoveries section so they are not smuggled into the proposal's scope.
Result
Lead with the two verdicts: adopt, adopt with changes, or reject, followed by a compact rationale. Then present only the sections supported by evidence:
- claim-verification ledger;
- consumer, dependency, and blast-radius map;
- adjacent-mechanism dispositions;
- engineering-principle fit;
- alternatives and frontier judgment;
- recommended delivery shape;
- owner decisions; and
- standalone discoveries.
Mark inference and uncertainty explicitly. Prefer exact evidence and concise decision-ready prose over ceremony or a fixed report length.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

