selfloop

A command for starting and controlling a persistent loop that improves SIPS and the agent through measured, verified iterations. It can start, pause, resume, inspect, complete, or clear the loop.

In plain words
What is it for?
Running repeated self-improvement cycles, checking their status, pausing or resuming them, and ending a loop when its objective is genuinely complete.
Why use it?
Ongoing improvement needs a recorded objective and explicit state rather than disconnected experiments. This command keeps the loop’s progress and controls in one place.

Command

Install

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.

agentmods
npx agentmods add commands/rasputinkaiser/self-improvement-plugin/selfloop
Clone the repo
git clone --depth 1 https://github.com/RasputinKaiser/Self-Improvement-Plugin
Per session 29 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 751 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00029 $0.00751
Opus 5 $0.00015 $0.00376
Sonnet 5 $0.00006 $0.00150
Haiku 4.5 $0.00003 $0.00075

Measured 2d ago against content hash abd41bbf619a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

selfloop 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.

commands/selfloop.md · 75 lines

How it starts

The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/selfloop — evidence-driven self-improvement loop

Parse the user's arguments:

  • /selfloop [focus] starts a new self-improvement loop. focus is optional.
  • /selfloop status prints the current goal state.
  • /selfloop pause and /selfloop resume control the active loop.
  • /selfloop complete marks a genuinely finished loop complete.
  • /selfloop stop or /selfloop clear clears the goal and stops immediately.

Use ${PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/goal_state.py for every state transition. To start, run selfloop-set with the optional focus. For controls, use the existing status, pause, resume, complete, and clear actions. Immediately begin the first cycle after selfloop-set; do not wait for another message.

Progress surface

For an attached runtime campaign, read the Goal Board:

python3 scripts/sips_runtime.py read --op board --json '{"run_id":"<run_id>"}'

Poll with since_revision to receive bounded event deltas. The board exposes one foreground next action for a responsive operator surface while the runtime may still have multiple bounded workers. Its labels are presentation aliases; event digests, leases, and immutable receipts remain authoritative.

Scope lock

The loop's entire focus is improving SIPS or the agent operating it. Valid targets include reasoning quality, tool reliability, Memory Fabric recall, verification quality, context efficiency, autonomy, self-correction, and the proof surfaces that measure those capabilities. Do not drift into unrelated product work, cosmetic churn, score gaming, or changes without a measurable benefit.

One cycle

  1. Inspect current goal state and recall prior lessons for the relevant scope.
  2. Establish a baseline with SIPS status, self_correct.py --json, the most relevant tests, and any direct runtime proof needed for the target.
  3. Rank evidence-backed weaknesses by expected capability gain, confidence, recurrence, and implementation cost. Choose one. If uncertainty could change the choice, verify it before editing.
  4. State the candidate, baseline, expected gain, and acceptance check in one compact note.
  5. Run /checkpoint, then implement the smallest reversible change that can produce the gain.
  6. Run focused verification first, then the relevant SIPS regression checks. Compare against the baseline. Keep the change only when the evidence shows an improvement; otherwise repair it or restore the checkpoint.
  7. Record the cycle with goal_state.py selfloop-record improved "<summary and proof>", append the result to the SIPS improvement ledger, and use /teach for any durable lesson.
  8. Increment the goal turn count and immediately begin the next cycle while the goal remains active.

Read the full file on GitHub · 75 lines

Changes

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.

  1. 2d ago First seen · 75 lines · 29 tokens per session scan A abd41bbf619a

Subscribe to this mod's changes

selfloop is a command published in the GitHub repository RasputinKaiser/Self-Improvement-Plugin (6 stars, last pushed 4d ago), licensed MIT. It adds 29 tokens to every session and 751 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.