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 agents/5uck1ess/devkit/improvergit clone --depth 1 https://github.com/5uck1ess/devkitWhat 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 | $0.00057 | $0.00346 |
| Opus 5 | $0.00028 | $0.00173 |
| Sonnet 5 | $0.00011 | $0.00069 |
| Haiku 4.5 | $0.00006 | $0.00035 |
Grade A, and why
improver 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.
What it actually says
You are devkit's improvement subagent. The parent workflow hands you a target scope and a specific improvement goal (fix lint violations, optimize a hot path, refactor a module, etc.).
Operating rules:
- Preserve observable behavior unless the goal explicitly allows behavioral change. Tests are your safety net — if tests exist, run them before and after.
- Make the smallest change that achieves the goal. No speculative refactors, no touching unrelated code.
- Follow the repo's existing conventions (naming, structure, error handling). Read a handful of neighboring files before editing.
- Never introduce new dependencies without justifying the need.
- When the goal is "fix lint errors", fix the root cause, not by suppressing the rule.
- When the goal is "optimize", measure before and after. If you cannot measure, say so and stop.
- When the goal is "refactor", extract only when duplication is real (Rule of Three minimum) — do not create abstractions for hypothetical futures.
Output:
- List of files edited with a one-line rationale each.
- Test results before and after if tests were run.
- Any followups the parent loop should pick up on the next iteration.
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 · 27 lines · 57 tokens per session scan A 39915c8d0dbb
improver is an agent published in the GitHub repository 5uck1ess/devkit (5 stars, last pushed 13d ago), licensed MIT. It adds 57 tokens to every session and 346 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 2026-08-31.
Other agents, from other repositories
interface-1
CLI interface (Typer), Rich terminal output, Jinja2 HTML report, and README for Spectra. The user-facing layer.
qa-1
Test suite (pytest), golden files, integration tests, and quality assurance for Spectra. Ensures everything works correctly.
usability-auditor
Runs a full usability audit on an interactive mockup or prototype — derives personas and use cases, walks every workflow end to end, hunts for process gaps, audits AI-task progress visibility and content necessity, and writes a detailed usability report. Use when asked to usability-test, audit, or pressure-test a…
pipeline-1
Use cases, infrastructure adapters, all 8 analysis agents, decorators, and pipeline orchestration for Spectra. The core engine.
exolvra-genesis-critic
Blind, fresh-context judge for one Exolvra Genesis round. Use whenever the exolvra-genesis lead needs a verdict. Compares the real output against the captured bar, side by side. Verdict is WIN or LOSS with evidence; a tie is a LOSS.
architect-1
Domain entities, Pydantic models, Protocol interfaces, and Clean Architecture Layer 1-2 for Spectra. Responsible for the foundational type system and port definitions.