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 YiShu5/claude-skills --skill self-improving-agentgit clone --depth 1 https://github.com/YiShu5/claude-skillsWrote 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/yishu5/claude-skills/self-improving-agent)<a href="https://agentmods.dev/skills/yishu5/claude-skills/self-improving-agent"><img src="https://agentmods.dev/badge/skills/yishu5/claude-skills/self-improving-agent/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/yishu5/claude-skills/self-improving-agent"><img src="https://agentmods.dev/badge/skills/yishu5/claude-skills/self-improving-agent.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 | $0.00068 | $0.01127 |
| Opus 5 | $0.00034 | $0.00563 |
| Sonnet 5 | $0.00014 | $0.00225 |
| Haiku 4.5 | $0.00007 | $0.00113 |
Grade A, and why
self-improving-agent 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.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Self-Improving Agent (Low-Noise Edition)
A conservative self-improvement system that treats .learnings/ as a write-first
cache layer. Nothing reaches the main system files without human review.
Quick Reference
| Action | Target | Auto? |
|---|---|---|
| Log a learning | .learnings/LEARNINGS.md |
Yes |
| Log an error | .learnings/ERRORS.md |
Yes |
| Log a feature gap | .learnings/FEATURE_REQUESTS.md |
Yes |
| Promote to MEMORY.md | Requires user approval | No |
| Promote to SOUL.md | Requires user approval + review | No |
| Promote to AGENTS.md | Requires user approval | No |
| Promote to TOOLS.md | Requires user approval | No |
Core Principles
- Write to cache first. All observations go to
.learnings/. No exceptions. - Never auto-promote. The cache is not the system. Promotion is a deliberate act.
- SOUL.md is sacred. Modifications require explicit human review every time.
- MEMORY.md is for stable truths. Only long-term preferences, decisions, and goals belong there.
- Errors and learnings stay separate. Different logs, different purposes.
- Silence is better than noise. If unsure whether something is worth logging, don't log it.
- No cross-session broadcasting of raw learnings. Only promote distilled, verified content.
Logging Format
Learning Entry
## [YYYY-MM-DD] Category: brief title
- **Type**: correction | knowledge_gap | best_practice
- **Context**: What was happening
- **Learning**: What was learned
- **Confidence**: low | medium | high
- **Promote?**: no (default) | candidate | promoted
Error Entry
## [YYYY-MM-DD] brief description
- **Command/Tool**: What failed
- **Error**: The error message or symptom
- **Root Cause**: What actually went wrong
- **Resolution**: How it was fixed
- **Recurrence**: first | repeated
Feature Request Entry
## [YYYY-MM-DD] brief description
- **Gap**: What capability is missing
- **User Need**: Why it matters
- **Priority**: low | medium | high
What ships with it
6 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.
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 · 133 lines · 68 tokens per session scan A ec8875185649
self-improving-agent is a skill published in the GitHub repository YiShu5/claude-skills (83 stars, last pushed 3d ago), licensed MIT. It adds 68 tokens to every session and 1,127 once invoked, about $0.0003 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
domain-memory
The Domain Memory pattern enables agents to work effectively across sessions without shared context. All state lives on DISK in domain-memory.yaml. Agents read from disk, work, write to disk, and exit.
pre-compact
Pre-compaction housekeeping. Walks a checklist (persistent memory updates, git hygiene, trash cleanup) plus an open-judgment audit, produces an SBAR with a go/no-go recommendation, and emits a copy-pasteable resume prompt for the post-compaction agent if work remains. Run this immediately before /compact.
agent-self-scheduling
Schedule AI agent runs with cron, loops, or external clocks while avoiding unsafe tight autonomous timers.
ai-agents-architect
Expert in designing and building autonomous AI agents. Masters tool use, memory systems, planning strategies, and multi-agent orchestration.
hive.note-taking
Maintain a free-form scratchpad of decisions, extracted values, and open questions so context pruning doesn't lose anything you still need.
notes
Skill "notes" from Pinvou/pinvou-agent, covering ima notes, operations, write rules, examples and response handling.