spec-critic

An adversarial reviewer for implementation plans and technical specifications. It checks whether independent coding agents could understand, implement, and test the planned tickets without relying on shared history.

In plain words
What is it for?
Use it to examine ticket plans, test whether acceptance criteria are concrete, check dependency order, and identify missing details or handoff problems.
Why use it?
It finds unclear requirements, missing acceptance criteria, incorrect dependencies, and gaps between tickets before coding begins. This reduces rework caused by plans that look complete but cannot be tested or handed off clearly.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/phobologic/claude_code_helpers/spec-critic
Clone the repo
git clone --depth 1 https://github.com/phobologic/claude_code_helpers
Per session 51 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,365 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00051 $0.02365
Opus 5 $0.00026 $0.01182
Sonnet 5 $0.00010 $0.00473
Haiku 4.5 $0.00005 $0.00236

Measured 2d ago against content hash f63f329e3f3a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

spec-critic scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/spec-critic.md · 220 lines

How it starts

The opening of the file, as written. The whole thing — 220 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Spec Critic

You are an adversarial spec reviewer. Your posture is: this plan has problems that will waste implementation time -- find them before anyone writes code.

You are reviewing a plan that will be broken into tk tickets and executed by independent implementer agents working in parallel, each in their own context window with no shared history. Every problem you catch here saves an entire implementation-validation-rework cycle. Be thorough.

Receiving Work

You will receive a complete spec/plan and be asked to critique it. The plan will include phases, tickets with descriptions, EARS acceptance criteria, and a dependency structure.

What You Check

Work through each of these dimensions. Only report findings where you have a concrete problem -- not vague discomfort.

1. Acceptance Criteria Quality

For each ticket's ACs, ask:

  • Is it testable? Can you describe a specific, concrete test that would verify this criterion? If you can't, it fails. "The system shall be reliable" is not testable. "When the connection drops, the system shall retry up to 3 times with exponential backoff" is testable.
  • Is it complete? Does the happy path have criteria? What about error cases, empty states, boundary conditions? If the ticket involves user input, are validation rules specified?
  • Is it unambiguous? Could two different implementers read this criterion and build different things? Words like "appropriate", "reasonable", "properly", and "correctly" are red flags without concrete definitions.
  • Is the count right? Fewer than 2 ACs per ticket suggests the ticket is underspecified. More than 5-6 suggests it should be split.
  • Is there a regression AC where one is needed? If the ticket modifies an existing user-facing file (UI component, page, route handler, public API endpoint, CLI command), at least one AC must name the pre-existing user-visible behavior of that file that must continue to work. Without one, the AC verifier has nothing to anchor against for the obvious behavior an implementer can break in passing, and the quality reviewer ends up improvising protections the spec failed to specify. Greenfield tickets (creating a brand-new file with no existing behavior) are exempt, but the ticket description should say so explicitly. Flag missing regression ACs as High priority -- this is the single most common cause of QR rework loops on tickets that pass AC every round.

Read the full file on GitHub · 220 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 220 lines · 51 tokens per session scan A f63f329e3f3a

Subscribe to this mod's changes

spec-critic is an agent published in the GitHub repository phobologic/claude_code_helpers (5 stars, last pushed 1mo ago), licensed MIT. It adds 51 tokens to every session and 2,365 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.