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 commands/benmarte/autoimprove/continuegit clone --depth 1 https://github.com/benmarte/autoimproveWhat 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 | $0.00012 | $0.00921 |
| Opus 5 | $0.00006 | $0.00461 |
| Sonnet 5 | $0.00002 | $0.00184 |
| Haiku 4.5 | $0.00001 | $0.00092 |
Grade A, and why
continue 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 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.
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 — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Resume an interrupted autoimprove session by reading the log.
Parse $ARGUMENTS:
- If a number is present, use it as the iteration count override. Otherwise use the remaining count from the log.
- If a quoted string is present, use it as the FOCUS override. Otherwise inherit the focus from the interrupted session.
Step 1: Read the log
Read .claude/autoimprove/log.md from the project root.
If the file doesn't exist, stop: "No previous session found. Use /autoimprove:improve to start one."
Step 2: Find the last session
Find the last ## Session — header in the log. If none found (legacy log format), stop: "No resumable session found. The log uses an older format without session headers. Use /autoimprove:improve to start a new session with the updated format."
Step 3: Check session status
Parse the **Status:** line from the last session header.
- If
COMPLETED→ stop: "Last session completed successfully. Use/autoimprove:improveto start a new one." - If
IN_PROGRESSor any other non-COMPLETED value → this session was interrupted. Proceed.
Step 4: Parse session metadata
Extract from the session header:
Planned: total planned iterations (number after "Planned:")Focus: the focus string (text between quotes after "Focus:", or "all improvement areas")Base commit: the commit SHA (text after "Base commit:")Baseline: the baseline score (number before "/100" after "Baseline:")
Step 5: Count completed iterations
Count the number of ## Iteration or ### Iteration headers that appear AFTER the last ## Session — header. This is the completed count.
Step 6: Calculate remaining iterations
remaining = Planned - completed- If the user provided N as an argument, use N instead of
remaining new_total = completed + remaining(orcompleted + Nif overridden)start_at = completed + 1
If remaining ≤ 0 and no N override: stop: "All planned iterations appear to have completed. Use /autoimprove:improve to start a new session."
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 · 98 lines · 12 tokens per session scan A f59a5ab4d4e3
continue is a command published in the GitHub repository benmarte/autoimprove (5 stars, last pushed 5mo ago), licensed MIT. It adds 12 tokens to every session and 921 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
crossframe-essay
使用 CrossFrame Essay 把用户给出的主题、素材或问题写成中文批判性洞察文章。.
crossframe-max
Use CrossFrame Max only when the user explicitly asks for maximum CrossFrame mode, exhaustive structural projection, full-scale world-model interpretation, or no word-limit complete explanation.
crossframe-suite
Use this command when a CrossFrame task needs multiple sibling skills in sequence.
crossframe-promax
PROMAX-NAMED-ONLY：只有 crossframe-promax、CrossFrame ProMax、$crossframe-promax、/crossframe-promax 四种明确名称具有触发权。.
crossframe-history
Use CrossFrame History for historical materials, civilization continuity, source-ledger boundaries, dynasty or institutional history, archive-access backlog, FOIA closeout, or local C5 boundary summaries.
crossframe-inquiry
Use CrossFrame Inquiry after a completed CrossFrame diagnosis, article, review, or suite workflow when the user wants follow-up questions, counterexamples, missing evidence, transfer conditions, or action boundaries.