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/binaryhb0916/isparto/claude-mdgit clone --depth 1 https://github.com/BinaryHB0916/iSpartoWhat 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 | $0.03640 | $0.03640 |
| Opus 5 | $0.01820 | $0.01820 |
| Sonnet 5 | $0.00728 | $0.00728 |
| Haiku 4.5 | $0.00364 | $0.00364 |
Grade A, and why
iSparto CLAUDE.md 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 3d 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 — 129 lines — stays where its author put it; the contents beside it link to each section on GitHub.
iSparto
Project Overview
iSparto is a cross-provider AI Agent Team workflow framework, built on Claude Code + Codex CLI, that turns a single coding agent into a team with distinct roles (Lead + Teammate + Developer + Doc Engineer + Process Observer + Independent Reviewer) — Claude reviews what GPT implements, and a zero-context GPT pass blind-reviews product-technical alignment. Target users are independent developers. Current stage: open-source core workflow released, dogfooding in progress.
Tech Stack
- Language: Shell (Bash), Markdown
- Framework: None (pure configuration project, driven by Claude Code slash commands + MCP)
- Platform: macOS (iTerm2 + tmux 3.x — tmux required since v0.8.0 for Independent Reviewer's
codex execinvocation in a tmux pane) - Build: No build step
- Other: Codex MCP Server (npx codex-mcp-server)
Development Rules
- Communicate and generate documentation in the user's language (English or Chinese only)
- Any code/command change must synchronously update the corresponding documentation (README, docs/, command header comments)
- Product direction changes must be written into documentation, not just discussed in conversation
- Ask me first about uncertain product questions; do not decide on your own
- plan.md update cadence: update
docs/plan.mdeither per-task (in the same commit as the task work) OR per-Wave (in the T10/close-out commit that lists all task completions with commit hashes — the Wave-close approach is acceptable when the Wave runs as a single atomic work session on a dedicated branch). Wave-completion entries and cross-session BLOCKING markers are written by/end-ispartoas part of the commit it generates, because that is the step that knows the Wave is fully complete. If a fix session does not correspond to any plan.md entry (e.g., a bug fix not tied to any Wave), no plan.md update is required. - plan.md verification-count accuracy: when a Wave completion entry records a commit count, compute it mechanically via
git log --oneline --no-merges <wave-base>..HEAD | wc -l— not by estimation. For entries authored pre-commit (standard/end-ispartocadence, where the Wave entry ships inside the same commit it documents), write the projected count and re-verify via the same command immediately after the commit lands; if mismatch, amend before push. Applies to every Wave close-out. The<wave-base>is the commit where the Wave branch diverged from the target branch. - Single TODO source (plan.md Backlog is authoritative): All TODOs, deferred actions, framework rule-fix candidates, and open commitments — including Process Observer audit findings about the framework itself, user-direction commitments surfaced in conversation, and Wave-internal deferrals — must be written to
docs/plan.md's Backlog section. Writing new backlog-type content to any other file is a workflow violation. Specifically forbidden channels: (a) creatingdocs/framework-feedback-*.mdfiles (pattern retired 2026-04-17); (b) embedding deferred items in Wave-entry prose (e.g., "Out of scope" paragraphs) without a parallel Backlog row; (c) storing project-level work items inmemory/— memory is for awareness information (user preferences, project facts, external references) only, not work; (d) listing "next-session to-dos" only indocs/session-log.mdNotes without a Backlog row./end-ispartoperforms a TODO-homing audit that blocks the commit if any new backlog-type content lands outsidedocs/plan.md. - Do not develop directly on main; feat/ for new features, fix/ for bug fixes, hotfix/ for urgent fixes, docs/ for pure documentation commits, release/ for releases
- install.sh changes must remain backward compatible (existing users must still be able to uninstall)
- Changes to command templates (commands/*.md) must be verified not to break existing users' /migrate-isparto and /init-isparto flows
- After completing all reviews, automatically create PR and merge to main — no manual user review needed
- Releases must use the
/release-ispartocommand — manualgit tag,git push origin <tag>, or any operation on main is not allowed. The release flow is fully automated byscripts/release.sh - This project is the framework itself; all Tier 1 System Prompt Layer files (as defined in Documentation Language Convention) fall within the self-referential boundary — Lead edits directly, and Process Observer interceptions can be approved. This includes both subdirectory files (
commands/,templates/,scripts/,hooks/,agents/,docs/,lib/,.github/) and root-level files (CLAUDE.md,CLAUDE-TEMPLATE.md,bootstrap.sh,install.sh,isparto.sh). Tier 2/3/4 documentation (otherdocs/*.md,README*.md,CONTRIBUTING.md,CHANGELOG.md,VERSION) is also in scope for direct Lead edits under the same framework self-referential principle.
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.
- 3d ago First seen · 129 lines · 3,640 tokens per session scan A 7bd0e28422c3
iSparto CLAUDE.md is an instructions file published in the GitHub repository BinaryHB0916/iSparto (59 stars, last pushed 2mo ago), licensed MIT. It adds 3,640 tokens to every session, about $0.0182 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 instructions, from other repositories
airflow CLAUDE.md
Claude Code instructions for apache/airflow, a project described as: Apache Airflow - A platform to programmatically author, schedule, and monitor workflows.
bytechef AGENTS.md
AGENTS.md instructions for bytechefhq/bytechef, covering bytechef agent instructions, project overview, code style & formatting, formatting rules and naming conventions.
TradingGoose-Studio CLAUDE.md
Claude Code instructions for TradingGoose/TradingGoose-Studio, covering claude.md, git rules and protected files.
superpowers-graph CLAUDE.md
Claude Code instructions for RonMizrahi/superpowers-graph, covering superpowers-graph — repository guide, sources of truth, changing the graph — the eval ships with the change and rules that are not obvious from the code.
Codex-Engineering-Workflow-Pack AGENTS.md
AGENTS.md instructions for SetraTheXX/Codex-Engineering-Workflow-Pack, a project described as: Local-first engineering workflow toolkit for Codex: skills, Coordinator Mode, worktree isolation, guarded dispatch, and parallel agent workflows.
workflows-mcp-server AGENTS.md
Instructions for cyanheads/workflows-mcp-server, covering agent protocol, what's next?, core rules, patterns and tool.