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 skills/indexnetwork/index/code-flow-artifactnpx skills add indexnetwork/index --skill code-flow-artifactgit clone --depth 1 https://github.com/indexnetwork/indexWhat 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 | $0.00181 | $0.01837 |
| Opus 5 | $0.00090 | $0.00919 |
| Sonnet 5 | $0.00036 | $0.00367 |
| Haiku 4.5 | $0.00018 | $0.00184 |
Grade A, and why
code-flow-artifact 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 — 142 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Code flow, drawn
The deliverable is a page a colleague can read cold and come away knowing how the thing works — including the part they could not have gotten by reading the files themselves. Two halves, and both have to be real: the tracing has to be honest, and the drawing has to show mechanism rather than vocabulary.
1. Trace before you write a word
Never describe a flow from memory or from names. Names lie — a function called
prepNode turned out to fetch data only one later stage uses, and a "clarifier"
turned out to be two model calls, not one. Both facts changed the conclusion.
Work outward from the entry points:
- Grep for the user-facing surface first: the tool name, the route, the CLI verb,
the button's handler.
grep -rn "create_intent\|/intents/confirm"finds doors. - Read the graph/router wiring next if there is one. It hands you the node list and the conditional edges — that is the flow's skeleton for free.
- Then read each node. Read the prompts and schemas, not just the code around them. In model-backed systems the semantics live in the prompt: which speech acts get dropped, what the model is forbidden to invent, what a null means.
- Follow what gets enqueued. Background work after the main write is part of the flow, and it is the part people forget they own.
- Check what actually runs, not what the file defaults to. Config defaults are often the dead branch; look for the env/flag resolver and say which value is live.
Cite file.ts:123 for every non-obvious claim. It makes the page auditable, and
it is what turns "a doc" into "a reference".
2. Find the shape
Most real flows are not a line. Before drawing, name the shape out loud:
- Doors → spine → tail is the most common: several entry points converge on a shared core, and a queue picks up afterward.
- Fork at a gate: one decision point with two or three genuinely different exits, each of which deserves its own path.
- Loop until settled: a retry, a reconciliation, a budget being spent down.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 · 142 lines · 181 tokens per session scan A 309865e05766
code-flow-artifact is a skill published in the GitHub repository indexnetwork/index (132 stars, last pushed 3d ago), licensed MIT. It adds 181 tokens to every session and 1,837 once invoked, about $0.0009 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 2026-08-30.
Other skills, from other repositories
autovault-skill
Understand AutoVault-managed skills and how to install or update them. Use when a skill is visible via symlink, when authoring or editing SKILL.md, or before touching /.autovault/skills. Vault writes must go through autovault add-local or MCP proposeskill/updateskill so AutoVault re-signs; never hand-edit vaulted…
using-geoscience-skills
Foundation skill for the geoscience skills library. Routes user intent to the correct domain skills, slash commands, and workflow skills. Loaded automatically at session start via SessionStart hook.
ha-discovery-augment
Augments an existing Home Assistant Custom Integration with one network or bus discovery mechanism beyond Zeroconf — DHCP, SSDP/uPnP, USB, HomeKit, or MQTT discovery — conforming to spec/ha/discovery-mechanisms. Sets the manifest matcher list, implements the typed asyncstep in configflow.py, forwards into a confirm…
discovery-distribution
Run the discovery → synthesis → build → verify → elite-package → distribute loop to turn domain insight into a multi-surface, verified GitHub product. Use when productizing an idea, packaging agentic systems, or raising a repo to elite open-source standards before launch.
agent-code-analyzer
Agent skill for code-analyzer - invoke with $agent-code-analyzer.
foundry-config-setup
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded projectendpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.