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 sungurerdim/dev-skills --skill ds-tunegit clone --depth 1 https://github.com/sungurerdim/dev-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/sungurerdim/dev-skills/ds-tune)<a href="https://agentmods.dev/skills/sungurerdim/dev-skills/ds-tune"><img src="https://agentmods.dev/badge/skills/sungurerdim/dev-skills/ds-tune.svg" alt="Measured on agentmods" 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 | $0.00034 | $0.04298 |
| Opus 5 | $0.00017 | $0.02149 |
| Sonnet 5 | $0.00007 | $0.00860 |
| Haiku 4.5 | $0.00003 | $0.00430 |
Grade A, and why
ds-tune 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 4d 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 — 208 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/ds-tune
Manual optimization is slow — 8-10 experiments per day, subjective judgment, no audit trail. Skill runs 100+ experiments autonomously, keeping only what measurably improves.
Autonomous Optimization — Karpathy's autoresearch pattern generalized for any project with a measurable metric. Thanks to Andrej Karpathy for open-sourcing the core idea.
Completion Evidence — applies to every phase: Report
done/OKonly with the machine-checkable evidence the gates name — the exact command run and its observed output (orfile:linediff). Missing evidence → reportINCOMPLETEplus what is missing. Self-assessment is never evidence. (This band repeats at file end by design — both copies are normative.)
Triggers
- User runs
/ds-tune - User asks to optimize, tune, or improve a measurable aspect of their project
- User asks "make this faster", "improve accuracy", "reduce bundle size", or similar
- User asks to set up a self-improving loop
Triggers — INVOKE / DON'T INVOKE
| INVOKE | DON'T INVOKE |
|---|---|
| "make this faster", "optimize bundle size", "self-improving loop" | "audit performance once (no loop)" (→ ds-review --perf) |
| "100+ experiments for one metric" | "set performance budget + CI gate" (→ ds-launch --perf-budget) |
| "Karpathy-style autoresearch loop" | "research optimization techniques abstractly" (→ ds-research) |
| "git-ratchet: keep only improvements" | "manual optimization decisions" (→ user owns the choice) |
Contract
Dimensions: D1
- One file, one metric, one loop — Karpathy's core constraint. Skill generates optimization infrastructure (
ds/tune/), then runs the loop under the git ratchet defined in Quality Gates (only measured improvements survive; every experiment committed before mechanical evaluation). - Standalone: use blueprint when available; own analysis when absent.
- Full accounting enforced: every finding and planned check ends in an explicit disposition (fixed / skipped + reason / only you can do); summary totals balance.
- Pre-existing / out-of-scope errors detected during work are NOT skipped — fixed inline or escalated with concrete blocker.
What ships with it
5 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.
- 4d ago Changed · -43 lines 4dc3331a684b
- 7d ago First seen · 251 lines · 34 tokens per session scan A 90c334758ee7
ds-tune is a skill published in the GitHub repository sungurerdim/dev-skills (1 stars, last pushed 2d ago), licensed MIT. It adds 34 tokens to every session and 4,298 once invoked, about $0.0002 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 skills, from other repositories
ring:migrating-to-lib-observability
Migrating a Lerian Go app off lib-commons observability imports (deprecated shims or removed APIs) to lib-observability via a fixed mapping table, then bumps go.mod and validates the build; ring:backend-go applies the edits. Covers log/zap/runtime/assert, opentelemetry/tracing, HTTP middleware, context helpers, and…
ring:using-lib-streaming
Using lib-streaming, Lerian's producer-only event publication library (Kafka/SQS/RabbitMQ/EventBridge), in two modes. Sweep Mode detects DIY publishers (franz-go, sarama, amqp091, watermill, raw AWS SDK) and re-rolled manifests/breakers. Reference Mode catalogs the Builder/Emitter/Catalog facade. Companion to…
ring:using-lib-systemplane
Using lib-systemplane, the hot-reload runtime-config plane (Postgres LISTEN/NOTIFY or MongoDB change streams), in two modes. Sweep Mode detects DIY config reload (SIGHUP, fsnotify, viper, pgx LISTEN), manual tenant-scoping, hand-built admin CRUD, and v4 residue. Reference Mode catalogs client lifecycle and…
ring:applying-composition-patterns
React composition patterns that scale. Avoid boolean prop proliferation by using compound components, lifting state, and composing internals. Use when refactoring components with boolean prop proliferation, building flexible component libraries, or during architecture review. Skip for simple components with 1-2 props…
ring:using-lib-observability
Using lib-observability v1.1.0, Lerian's OpenTelemetry foundation (lib-commons, lib-systemplane, lib-streaming depend on it), in two modes. Sweep Mode detects DIY zap/slog logging, raw OTel metrics, hand-rolled redaction, and hard-coded attribute strings. Reference Mode catalogs the log, metrics, zap, redaction, and…
ring:using-outbox
Using the transactional-outbox pattern across lib-streaming (writer) and lib-commons/v5/commons/outbox (repository + relay), in two modes. Sweep Mode detects DIY outbox tables, hand-rolled relay loops, send-and-pray emits, missing WithOutboxTx wrapping, and broker calls inside DB transactions. Reference Mode catalogs…