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 HuntsDesk/ve-kit --skill self-improvegit clone --depth 1 https://github.com/HuntsDesk/ve-kitWrote 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/huntsdesk/ve-kit/self-improve)<a href="https://agentmods.dev/skills/huntsdesk/ve-kit/self-improve"><img src="https://agentmods.dev/badge/skills/huntsdesk/ve-kit/self-improve/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/huntsdesk/ve-kit/self-improve"><img src="https://agentmods.dev/badge/skills/huntsdesk/ve-kit/self-improve.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00088 | $0.02958 |
| Opus 5.5 | $0.00035 | $0.01183 |
| Sonnet 5.5 | $0.00018 | $0.00592 |
| Haiku 4.5 | $0.00009 | $0.00296 |
Grade A, and why
self-improve 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 11d 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 — 127 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Self-Improve — learn from review findings, encode guardrails
This skill closes the agent-fleet feedback loop: it reads what code-reviewers and batch retrospectives have been catching, finds the recurring root causes, and encodes them as guardrails so the same class of mistake stops recurring.
Owner: project-coordinator agent (the self-improvement domain + the consumer that runs it). Pairs with docs-manager for any agent/skill/rule edits (that's docs-manager's lifecycle domain — this skill proposes, docs-manager applies the config-layer ones).
How this differs from the review-* family (do NOT duplicate)
review-* (docs-manager) |
/self-improve (this skill) |
|
|---|---|---|
| Question | "Is the config well-formed + current?" | "What keeps going wrong, and how do we prevent it?" |
| Input | The .claude/ artifacts themselves |
Board review findings + batch outcomes + LESSONS.md |
| Output | Structural fixes (frontmatter, stale refs, model IDs) | New preventive rules (LESSONS.md, agent hardening, flagged proposals) |
| Trigger | Model upgrade, quarterly, pre-initiative | After ~5 batches, a finding cluster, or a notable incident |
If a finding is "the config is structurally stale" → that's a review-* job; route it there. This skill is for behavioral patterns (recurring bugs, repeated reviewer catches, repeated skips).
When to invoke
- After every ~5 worker batches (the auto-claude worker's CLOSING step may call this on a counter).
- When you notice a cluster of related review findings across recent sessions.
- After a notable incident whose root cause would recur without a guardrail.
- Periodically (monthly) as a fleet-health pass.
The one rule that governs everything here
A one-off is NOT a rule. Only encode a guardrail when the pattern recurred across ≥2 distinct batches/sessions/findings. A single incident gets a shared-memory note (cross-dev) or a board task, not a permanent rule. Over-encoding turns LESSONS.md + agent prompts into noise that gets skimmed and ignored. Be ruthless: the bar is "recurring root cause," not "something went wrong once."
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.
- 11d ago Changed 530c4828fb65
- 1mo ago Changed · +1 lines f4ef636c7a81
- 1mo ago Changed · +1 lines e8c9fd83c061
- 1mo ago Changed · +7 lines 91087738cb45
- 1mo ago Changed af9c19079f02
- 1mo ago First seen · 118 lines · 88 tokens per session scan A 22845497bbbd
self-improve is a skill published in the GitHub repository HuntsDesk/ve-kit (3 stars, last pushed 10d ago), licensed MIT. It adds 88 tokens to every session and 2,958 once invoked, about $0.0004 per session on Opus 5.5. 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 skills, from other repositories
pr-review
Review open PRs two ways - default is a per-PR deep review with severity-tagged findings, inline comments, and a verdict; --survey runs a risk-tiered triage digest of what's safe to merge first.
pr-triage
First-touch triage for external pull requests - verdict, label, and a welcoming comment within minutes of open.
feedback-builder
Point Aeon at a service's /feedback endpoint - pull what agents reported (bugs, missing features), triage and cluster it, then build the best accepted request as a PR for a human to approve.
[REPLACE: SKILL_NAME]
First-touch review of newly opened PRs on [REPLACE: WATCHEDREPO] — verdict + welcoming comment + label.
code-review
Systematic code review for bugs, security, style, and performance.
code-review
Review code for security, quality, and maintainability with structured feedback.