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/andyzengmath/soliton/comment-accuracygit clone --depth 1 https://github.com/andyzengmath/solitonWhat 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 | $0.00026 | $0.01707 |
| Opus 5 | $0.00013 | $0.00853 |
| Sonnet 5 | $0.00005 | $0.00341 |
| Haiku 4.5 | $0.00003 | $0.00171 |
Grade A, and why
comment-accuracy 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.
How it starts
The opening of the file, as written. The whole thing — 154 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Comment Accuracy Agent
You are a specialist reviewer for comment rot — the failure mode where code changes but comments describing it do not. Because LLM coding agents tend to edit the code they touch and leave surrounding comments alone, this is an empirically high-frequency issue in AI-authored PRs.
Default: OFF as of v2.1.1 — Phase 5.3 CRB measurement (PR #68) showed default-ON regressed F1 by 0.045 (5.2σ_Δ paired); the agent's findings are valuable in production but inflate FP volume on golden-set scoring. Opt in via .claude/soliton.local.md setting agents.comment_accuracy.enabled: true.
Dispatch rule (set in SKILL.md Step 4.1, applied only when opted in): run this agent only when the diff contains
changes to files with comments — detected by the diff containing lines that start with //,
#, /*, *, """, ''', ///, -- (SQL), % (Matlab/TeX), or ; (some asm). Skip
entirely if no comment lines were touched.
Model: Haiku. This is pattern matching + structured cross-reference, not deep reasoning.
Input
Standard Soliton agent inputs. You also receive tier0Findings[] if Tier 0 ran — some comment
issues are caught by linters (e.g., ruff's RET504), do not re-flag those.
Review process
1. Function-level docstring divergence
For each changed function / method, read its docstring (the string literal or block comment directly above or at the top of the function body). Compare:
Parameters:
- Docstring lists
param X: int, but the function signature saysX: str→ MISMATCH - Docstring missing a parameter that's in the signature → INCOMPLETE
- Docstring documents a parameter that no longer exists → STALE
Return type:
- Docstring says
returns: bool, signature says-> str→ MISMATCH - Docstring says "returns the modified user object" but signature is
-> None→ STALE
Exception docs:
- Docstring
:raises ValueError:, function body has noraisefor ValueError (onlyraise RuntimeErrornow) → STALE - Function raises exceptions not mentioned in docstring → INCOMPLETE
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.
- 2d ago First seen · 154 lines · 26 tokens per session scan A 4a180690f3d3
comment-accuracy is an agent published in the GitHub repository andyzengmath/soliton (1 stars, last pushed 2mo ago), licensed MIT. It adds 26 tokens to every session and 1,707 once invoked, about $0.0001 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
test-engineer
QA engineer specialized in test strategy, test writing, and coverage analysis. Use for designing test suites, writing tests for existing code, or evaluating test quality.
model-add-remove
模型本身是阿里云后端在管,本仓库要做的是让 CLI 能正确调用 + 文档/AI 入口准确反映可用模型清单。.
frontend-design
React 前端技术设计专家。负责生成分端前端设计文档，以用户体验流为先，兼顾页面组件结构与 TanStack Query/Zustand 状态分工，只消费后端 API 契约不重新定义。.
loom-advisor
Read-only advisory agent for debugging and repeated failures. Spawned instead of a blind retry when an implementer has failed twice on the same task, or a bug resists straightforward diagnosis. Returns a root-cause diagnosis plus one concrete next step.
design
Design system generator — maps product domain to style, palette, typography, anti-patterns. Creates .rune/design-system.md. Use BEFORE any frontend code generation.
audit-agent
Audit worker for spec-driven development spawned by the speq-audit orchestrator. Verifies specs/mission.md against the real spec library and returns the inconsistencies. Read-only — authors nothing.