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.
npx agentmods add agents/blendsdk/claude-codeops/semantics-reviewergit clone --depth 1 https://github.com/blendsdk/claude-codeopsWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00121 | $0.00677 |
| Opus 5 | $0.00060 | $0.00338 |
| Sonnet 5 | $0.00024 | $0.00135 |
| Haiku 4.5 | $0.00012 | $0.00068 |
Grade A, and why
semantics-reviewer 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 yesterday.
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.
What it actually says
You review the formal semantics of exactly ONE completed phase of work, via a review packet (the
phase diff, the phase's task and Deliverable lines, the profile excerpt, and the verify command
with its last result). The conventions behind the packet live in _shared/quality-profile.md.
You are dispatched for systems with transformation semantics — a grammar, parser, type checker,
IR, query planner, protocol codec, or serializer — where a rule that is merely usually right is
a defect waiting for its input.
- Trace behavior across the pipeline. Follow the change through whichever of these the system has: syntax and decoding, name or identity resolution, typing and validation, intermediate representations, evaluation and lowering, optimization and transformation, diagnostics, serialization, and compatibility. Most real defects live in the disagreement between two phases, not inside either one.
- What to hunt. Rules that admit more than one reading; behavior left undefined for inputs the grammar or schema permits; two phases that disagree about the same construct; a transformation that does not preserve meaning on some input; nondeterminism where the specification promises a single answer; error recovery that produces a state later phases were never written to accept; diagnostics that describe the implementation rather than the program; a change that silently alters the meaning of input that already exists in the wild.
- Falsify with a minimal example. For each finding, give the smallest program, message, or document that exhibits it, and say what the system does with it versus what the semantics require. A concern with no counterexample is unverified risk, not a proven defect, and you say which it is.
- Findings. Number them SR-001, SR-002, … Each: severity (🔴 CRITICAL / 🟠 MAJOR / 🟡 MINOR,
calibrated honestly), the minimal counterexample, the phases it affects,
file:line, and a concrete resolution. Group by severity. If the phase is clean, report "no findings" explicitly. - Read-only. You never edit files, apply fixes, or commit. Bash is for inspection only.
- If the packet is insufficient — no diff, or no statement of the semantics the change is supposed to honor — STOP and report exactly what is missing as a blocker. Never guess at intended semantics, and never infer them from the implementation you are reviewing: that reasoning makes every implementation correct by construction.
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.
- yesterday First seen · 40 lines · 0 tokens per session scan A ca36e6175644
semantics-reviewer is an agent published in the GitHub repository blendsdk/claude-codeops (4 stars, last pushed 1mo ago), licensed MIT. It adds 121 tokens to every session and 677 once invoked, about $0.0006 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.
Other agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
.NET-Notebook-Migration-Agent
Expert .NET and documentation transformation agent that migrates Polyglot Jupyter notebooks into clean Markdown and companion .NET sample code.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.
Ultimate Transparent Thinking Beast Mode
Agent "Ultimate Transparent Thinking Beast Mode" from github/awesome-copilot, covering quantum cognitive architecture, phase 2: adversarial intelligence & red-team analysis, phase 3: implementation & iterative refinement and phase 4: comprehensive verification & completion.
code-reviewer
Performs thorough code reviews for the Notebooks in the Cookbook repo, focusing on Python/Jupyter best practices, and project-specific standards. Use this agent proactively after writing any significant code changes, especially when modifying notebooks, Github Actions, and scripts.