critic

Critiques ICEA specs, Tech Specs, and generated code before they are written to disk.

1|Updated Jun 27, 2026
One-click install
npx skills add https://github.com/vickysrawat/AI-Assisted-development --skill critic-vickysrawat
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: critic
Source: https://github.com/vickysrawat/AI-Assisted-development/tree/main/skills/critic
Command: npx skills add https://github.com/vickysrawat/AI-Assisted-development --skill critic-vickysrawat

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Generated artefacts — ICEA requirement drafts, Tech Specs, and implementation code — can carry structural gaps, scope creep, or hidden assumptions that only surface after they are committed. This Skill evaluates each artefact at the moment it is produced, while both the spec and the output are still in context, catching intent misalignment before anything touches disk. ## Core Features & Use Cases - Three critique modes: ICEA mode checks conformance, completeness, testability, B-series sensitivity coverage, and scope; TECH mode checks ICEA↔design traceability, coverage-matrix completeness, and D-option fidelity; CODE mode checks traceability, simplicity, rules compliance, decision transparency, and hidden assumptions. - Bounded auto-revise loop: On a REVISE verdict, the parent skill regenerates the artefact and re-runs the critic up to 2 times before writing; unresolved cases surface to the developer with ACCEPT AS-IS, GUIDE, or HALT options. - Automatic and standalone invocation: Fires inside icea-feature (Steps 5 and 8) and icea-implement (Step 4a), or runs independently via /critic icea or /critic code. - Use Case: While generating a feature for ADO-1234, the critic flags that AC-4 has no planned file in the Tech Spec's AC→File table and that the code defaults pagination to 50/page without flagging the assumption — both fixed before any file is written. ## Quick Start Ask the assistant to run /critic icea ADO-1234 to critique the ICEA draft for that work item before approving it.

Frequently Asked Questions about critic

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I critique an ICEA draft before approving it?

Run /critic icea ADO-<id> to critique the ICEA file for that work item against conformance, completeness, testability, B-series coverage, and scope dimensions. If no id is given, the skill resolves the ADO id from the current branch name.

How does the critic review generated code before it is written?

In code mode the critic compares generated code against the approved ICEA and Tech Spec for traceability, simplicity, rules compliance, decision transparency, and hidden assumptions. A REVISE verdict blocks the disk write and triggers regeneration, up to two automatic retries.

When does the critic run automatically in the ICEA workflow?

It fires inside icea-feature at Step 5 for the ICEA draft and Step 8 for the Tech Spec draft, each before the temp-file write, and inside icea-implement at Step 4a after code generation but before any disk write.

Does the critic modify or fix the artefacts it reviews?

No. The critic is ephemeral: it writes no ledger, assigns no fingerprints, and applies no fixes. Its only output is a verdict and a list of concrete concerns; regeneration is performed by the parent skill.

What happens if the critic still fails after automatic retries?

After two failed revisions the loop stops and surfaces the remaining concerns to the developer with three choices: ACCEPT AS-IS to write anyway, GUIDE to provide specific direction for one more attempt, or HALT to write nothing.

Can the critic read source files on disk?

Only in standalone code mode (/critic code), which is Category A consent and announces the file scope before reading. All internal gates and ICEA/tech modes are Category C and never read source files.