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 instructions/rushengzhou/sid-code/claude-mdgit clone --depth 1 https://github.com/rushengzhou/sid-codeWhat 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 | $0.09710 | $0.09710 |
| Opus 5 | $0.04855 | $0.04855 |
| Sonnet 5 | $0.01942 | $0.01942 |
| Haiku 4.5 | $0.00971 | $0.00971 |
Grade A, and why
sid-code CLAUDE.md scanned grade A with 1 finding 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.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -fsSL https://www.sid-code.cc/releases/sid-code/install.sh | SID_CODE_CHANNEL=beta bash How it starts
The opening of the file, as written. The whole thing — 414 lines — stays where its author put it; the contents beside it link to each section on GitHub.
sid-code — 长在企业研发环境里的 coding agent
北极星：长期宗旨与方向（最高纲领，长期不变）
一句话宗旨：别人给你一个 agent；sid-code 给你一个你能改、能量、能审、数据不出门的 agent 底座——然后用你自己的轨迹数据，让它每个版本都更快、更省、更少返工。
四大特性（能演示）
- 企业级 —— 长在企业研发环境里：内部网关（含计费取价口径）、内网 GitLab、企业 SSO、MCP、团队默认配置分发、企业 policy、缺陷系统闭环。用户收益是「装上就接得上你公司已有的那套东西，不用先改造企业来适配工具」。
- 可定制 —— harness 整套可改，模型任你换。三层：配置层（多 provider 三族协议、降级链、权限规则、团队默认值，改一行配置）/ 扩展层（Hook、Skill、子代理、MCP、插件多种源，写一个文件）/ 源码层（工具、上下文工程、主循环全部开源可改，提一个 PR）。
- 数据主权 —— 代码、对话、轨迹、评测、成本账本全部本地，不进任何人的训练集，数据在自己手上，才谈得上用它做优化。
- 可观测 —— 每一轮的耗时、成本、决策都留有轨迹，发布前能跑评测知道有没有退步。它不是技术细节，是四大方向的载体：没有轨迹就画不出那四条曲线。
四大方向（能复算 · 按 release 出一条曲线）
每个方向都是一个主口径（那条曲线）+ 一组辅助口径（曲线动了要能归因到哪一层）。 只有主口径进 release 曲线，辅助口径是排查时的分解项——但辅助口径缺失时主口径动了也说不清为什么， 所以两者都列。三档标注按事实分：✅ 轨迹里已有（可直接复算）/ ⚠️ 有采集但口径要小心 / ❌ 尚未采集（想用先补埋点，别拿它当结论）。
四条跨方向通用铁律（都是踩出来的）：
- 一律看 p95/p99，均值会骗人。慢尾巴才是用户流失点。
- 每个指标必须能指到源字段。说不出取数源的数字就是自我感觉，见自检第 2 问。
- 分母比分子重要。「命中率」「成功率」「触发率」的分母口径一变，曲线就整体平移——分母必须和指标一起写死。
- 区分 stock 与 flow。末次快照值（如
total_tokens_sent）除以累加值（如total_cost_usd）得到的是错数， 要用累积字段（total_cumulative_prompt_tokens）。
更快 —— 延迟类
| 层 | 指标 | 状态 / 取数源 |
|---|---|---|
| 主口径 | TTFT p50/p95/p99 + 端到端耗时（用户回车 → 最终答复） | ✅ StreamPhase(first_content).ttft_ms 是唯一干净源 |
| 归因：卡在哪一段 | TTFB（首字节，拆「网关握手」vs「模型 prefill」） | ⚠️ headers_received.ttfb_ms：协议层同口径，语义受网关缓冲策略影响，跨路由不可比——必须按 model 分组看（latencyByModel） |
| 纯生成耗时 gen p50/p95/p99（单次 fetch，不含重试） | ✅ RetryTelemetry(stream_completed).elapsedMs |
|
| 整轮 API 耗时（含握手 + 生成 + 重试） | ⚠️ AfterModelRaw.elapsed_ms，别与 TTFB 混，渲染必须标「整轮」 |
|
| 工具执行耗时 —— 回答「慢在模型还是慢在工具」 | ✅ PostToolUse.duration_ms / total_tool_duration_ms |
|
| 体感 | tokens/sec（输出流速，比 TTFT 更贴「生成快慢」体感） | ✅ output_tokens_per_sec |
| TTFT 按缓存命中分桶 —— 缓存到底让首字快了多少的唯一对照口径 | ✅ ttftByCache.hit/miss |
|
| 缺口 | TPOT / ITL（token 间隔，决定打字流畅感）、Goodput（满足 SLO 的有效吞吐） | ❌ 全仓零命中；goodput 还需先定义 SLO |
TTFT 的口径铁律（这条有 P0 bug 教训）：必须是首个任意内容 chunk（含 thinking / tool_use）， 且每次 fetch 单独计、不跨重试累计。只在可视文本上计 → 对 thinking 模型和纯工具调用轮系统性虚高数十秒。
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 · 414 lines · 9,710 tokens per session scan A c5c7f1cd07f6
sid-code CLAUDE.md is an instructions file published in the GitHub repository rushengzhou/sid-code (2 stars, last pushed 2d ago), licensed MIT. It adds 9,710 tokens to every session, about $0.0485 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
cicero AGENTS.md
Instructions for 5uck1ess/cicero, covering agent guidance, product boundary, toolchain, verification and reliability and security invariants.
opencode-dejavu AGENTS.md
Instructions for WhiteBite/opencode-dejavu, covering project knowledge base, structure, where to look, code map and conventions.
cicero CLAUDE.md
Instructions for 5uck1ess/cicero, a project described as: Self-hosted voice for coding agents. Talk from any browser or a Telegram call, interrupt mid-sentence, clone any voice, and hand real work to agents that ship PRs. Local STT/TTS; real-time voice client for ACP.
ob-1 AGENTS.md
Instructions for Overbrilliant/ob-1, covering agents.md, project index, stack, commands and key files.
vibe-codr AGENTS.md
Instructions for robzilla1738/vibe-codr, covering agents.md — vibe-codr, stack & layout, commands, visible tui change and conventions.
lfcode AGENTS.md
Instructions for lfyxhappy/lfcode, covering project structure, issue records, build and test, branch names and commits and pr titles.