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/lenamonj/jeffy-loop/jeffynpx skills add lenamonj/jeffy-loop --skill jeffygit clone --depth 1 https://github.com/lenamonj/jeffy-loopWhat 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 | $0.00025 | $0.05464 |
| Opus 5 | $0.00013 | $0.02732 |
| Sonnet 5 | $0.00005 | $0.01093 |
| Haiku 4.5 | $0.00003 | $0.00546 |
Grade E, and why
jeffy scanned grade E with 3 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Tells the agent never to refusehighAnti-refusal
Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.
8. Repository scope: if the project is a git repository, compare `git rev-parse --show-toplevel` with the project root. When they differ, the project is a subdirectory of a larger repository, so every checkpoint's `git a Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
1. Hook dependency: `which jq`. If missing the Stop hook cannot parse its input and the loop cannot run. Suggest `winget install jqlang.jq` (Windows), `brew install jq` (macOS), or `sudo apt-get install jq` (Debian/Ubunt Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
cat > "$PR/.claude/jeffy-loop.local.md" <<EOF How it starts
The opening of the file, as written. The whole thing — 149 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Jeffy
Bootstrap per-project Jeffy state files, write the loop state file, and begin iteration 1. The loop engine is Jeffy's own Stop hook, shipped with this skill at hooks/stop-hook.sh: at every turn end it reads .claude/jeffy-loop.local.md at the project root, and while that file names this session and budget remains it re-feeds the iteration prompt; it deletes the file and lets the session end when the budget is spent, the completion promise fires, or one of its own gates ends the run - a stall, an oscillation, a time ceiling, the sweep fail-fast, or an audit or refill inside the closing window. The hook anchors itself to CLAUDE_PROJECT_DIR, which hooks receive fixed at the directory Claude Code was started in, so shell cwd drift mid-iteration cannot kill the loop. The hook is registered machine-wide in ~/.claude/settings.json by the installer but exits immediately in any session whose project has no state file. This skill only sets up and launches; the hook implements the loop mechanics.
Project root means the directory Claude Code was started in - the session's primary working directory shown in the environment, which is stable regardless of shell drift. Resolve it to an absolute path with forward slashes and use it everywhere below; never trust the shell's current cwd, which persists across Bash calls and may have drifted into a subdirectory.
Arguments
Parse $ARGUMENTS: if the first token is an integer, it is the iteration budget N, default 10. Next parse the ceiling flags wherever they appear in the remainder, each taking one value: --max-time <duration> sets max_wall_clock_seconds, --max-iter-time <duration> sets max_iteration_seconds, and --max-context <N> sets max_context_growth; a duration is a bare number of seconds or a number suffixed s, m, or h (45m, 2h, 900s, 900), and 0 means off, the default for all three. A flag missing its value, an unparseable duration, or an unrecognized -- token is a refusal naming the token, never silently folded into the focus - a documented ceiling that becomes free text is a ceiling that is silently off. Whatever text remains after the budget and the flags is the focus directive for this run. If the token after the optional budget is exactly enhance, refuse the launch and stop: Enhance mode was removed in v1.11.0 - the defect loop is the product - and the last release that carries it is v1.10.0. On a fresh project, iteration 1 is always consumed by the audit that generates the backlog, so N=2 executes exactly one task; if the user picks N below 5, proceed but note that larger budgets make materially more progress.
What ships with it
9 files 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.
- hooks/lib/check-claims.sh 6.3 KB runs code
- hooks/lib/detect-sandbox.sh 1.7 KB runs code
- hooks/lib/quiet-verify.sh 11 KB runs code
- hooks/lib/run-probe.sh 4.3 KB runs code
- hooks/stop-hook.sh 142 KB runs code
- references/backlog-default.md 3.3 KB
- references/iteration-prompt.txt 31 KB
- references/journal-default.md 2.5 KB
- references/plan-default.md 37 KB
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 · 149 lines · 25 tokens per session scan E e8f5d39924f3
jeffy is a skill published in the GitHub repository lenamonj/jeffy-loop (8 stars, last pushed 3d ago), licensed MIT. It adds 25 tokens to every session and 5,464 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it E with 3 findings (tells the agent never to refuse, asks for root, reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
paper-illustration
Generate publication-quality AI illustrations for academic papers using Gemini image generation. Creates architecture diagrams, method illustrations with Claude-supervised iterative refinement loop. Use when user says "生成图表", "画架构图", "AI绘图", "paper illustration", "generate diagram", or needs visual figures for papers.
integrity-forensics
Run the Anti-Autoresearch integrity-forensics sweep (span-anchored evidence ledger → GPT auditors propose findings → a rules-only reporter that lists every proposal with what the auditor said about it) against a paper via a SHA-pinned thin launcher — then convert the verdict into a typed policy gate…
patent-novelty-check
Assess patent novelty and non-obviousness against prior art. Use when user says "专利查新", "patent novelty", "可专利性评估", "patentability check", or wants to evaluate if an invention is patentable.
overleaf-sync
Two-way sync between a local paper directory and an Overleaf project, so ARIS audit/edit workflows stay on the local copy while collaborators edit in the Overleaf web UI. Use when user says "同步 overleaf", "overleaf sync", "推送到 overleaf", "connect overleaf", "Overleaf 桥接", "pull overleaf", "push overleaf", or wants to…
flow-next-tracker-sync
Project a flow-next spec to a tracker issue (Linear, GitHub, GitLab, Jira) and reconcile two-way. Use when asked to sync to a tracker. NOT plan-sync.
flow-next-export-context
Export RepoPrompt context to a markdown file for review with an external LLM (ChatGPT, Claude web, etc.). Use when you want Carmack-level review but prefer an external model. Triggers on "export context", "export for external review", "export plan for ChatGPT", "export impl review context", "review with an external…