Kungfu is a work-continuity system that lets the same task move between coding agents such as Codex, Claude, and OpenCode without losing its context or progress. It is for people coordinating different agents across attempts, reviews, failures, and recoveries, while the catalogue skills and instructions define agent workflows around that shared work.
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 skills add kungfu-systems/kungfu --skill claudegit clone --depth 1 https://github.com/kungfu-systems/kungfuWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/kungfu-systems/kungfu/claude)<a href="https://agentmods.dev/skills/kungfu-systems/kungfu/claude"><img src="https://agentmods.dev/badge/skills/kungfu-systems/kungfu/claude/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/kungfu-systems/kungfu/claude"><img src="https://agentmods.dev/badge/skills/kungfu-systems/kungfu/claude.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What 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 5.1 | $0.00049 | $0.00643 |
| Opus 5 | $0.00024 | $0.00321 |
| Sonnet 5 | $0.00010 | $0.00129 |
| Haiku 4.5 | $0.00005 | $0.00064 |
Grade A, and why
kungfu-agent-onboarding scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
Kungfu Agent Onboarding
When KUNGFU_AGENT_ENVIRONMENT=native-interactive, treat the injected
Console/Skill/WorkRef envelopes as discovery pointers, never as prior chat or
completion proof. Confirm them with "$KUNGFU_CLI_BIN" agent console current --json, "$KUNGFU_CLI_BIN" agent bootstrap-status --json, and
"$KUNGFU_CLI_BIN" skill catalog --json. Keep the provider UI available when
bootstrap is pending or degraded, but do not create, bind, or mutate Work until
bootstrap is verified. Before the first Work
mutation, bind the chosen Assignment with "$KUNGFU_CLI_BIN" agent console bind-work --initiative-id <id> --assignment-id <id> --json; stop unless the
result is status: bound, including when another native writer is active.
Run kungfu agent brief, then kungfu agent docs --verify --json and
kungfu agent map --json. Select only the route relevant to the user's task.
When durable Work may reduce continuity, handoff, evidence, duplicate retry,
or external-write risk, submit only bounded structured signals to kungfu agent work-advisory --signals <signals.json> --json. Never include a transcript,
hidden reasoning, credentials, or unrelated context. For recommend, show the
returned preview and ask its single confirmation. Only after confirmation use
the returned existing kungfu.work.capture, kungfu.work.admit, and
kungfu.agent.console.bind-work path, cite its receipts, and continue the
original task. Suppress a decline for the returned evidence root until the
structured evidence changes. Advice grants no external authority.
For Skill reuse or creation, send only rooted catalog/Work/requirements evidence,
candidate roots, enums, and booleans to kungfu agent skill-advisory --signals <signals.json> --json. Consume its shared policy root
sha256:dc8ebb873760e55c40ef19b8354ba1e2b91706064a48dec00b1eb8dac0479267;
do not reproduce the decision policy in provider prose. The result is read-only.
Use kungfu agent context --task "<task>" --role <role> --budget <tokens> --route <route-id> --json when detail is needed. Stop on invalid roots,
ambiguity, stale state, or required omissions; use returned expansion handles
instead of loading the whole corpus.
Explain Kungfu in terms of what is already known about the user and workspace,
without claiming hidden knowledge. Offer one read-only or preview-first action.
Never infer authority from this Skill: writes require their public --execute
or authorization path, and Work completion requires native receipts.
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.
- 10d ago First seen · 47 lines · 49 tokens per session scan A 730164222fac
kungfu-agent-onboarding is a skill published in the GitHub repository kungfu-systems/kungfu (4,513 stars, last pushed today), licensed Apache-2.0. It adds 49 tokens to every session and 643 once invoked, about $0.0002 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
apple-tools
Reach Apple PIM data (Calendar, Contacts, Reminders, Messages, Location, Maps, Weather) from a pi session via iMCP + pi-mcp-adapter on macOS. Use when the user asks to read/search their Apple Calendar, Contacts, Reminders, iMessages, current Location, Maps, or Weather. Does NOT cover Apple Mail — iMCP exposes no Mail…
kb-setup
Set up, initialize, or configure the local markdown knowledge base for a project. Use when the user says "set up the knowledge base", "init the KB", "configure the markdown KB", "index my docs", "add a doc source to the KB", or wants the project's markdown searchable by kb search. Wraps kb init end-to-end through kb…
kb-search
Search the local markdown knowledge base for facts, prior decisions, definitions, people, error codes, config flags, and how-we-fixed-it notes. ALWAYS search here BEFORE answering a project-specific question from memory, guessing, or asking the user. Use whenever you hit an unknown term, an unfamiliar entity, an error…
ask-user
Ask the human structured questions via nui HITL instead of plain chat text.
efficient-execution
Reduce token waste and task completion time by choosing the simplest tool for each job.
author-pragma-dsl
Create and update validated pragma/v5 Expert, ExpertTeam, Flow, Evaluation, and Automation resources. Use when a user asks Pragma to create, change, configure, test, evaluate, or repair an Expert, team, Flow, Run Dry suite, schedule, trigger, or Automation in the current Pragma project.