A command that runs one cycle for improving a skill's output by benchmarking it, testing its writing style without revealing the treatment, reviewing evidence, and checking proposed engine changes.

Use it to run the self-improvement loop, optionally creating baseline comparisons or skipping selected stages, with results stored in timestamped learning-run folders.

It provides a repeatable way to find weaknesses and decide whether changes improve results.

improve — Command by Nisus74

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/nisus74/humanise/improve

Clone the repo

git clone --depth 1 https://github.com/Nisus74/humanise

Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 1,231 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.

Model	Per session	Once invoked
Fable 5	$0.00000	$0.01231
Opus 5	$0.00000	$0.00616
Sonnet 5	$0.00000	$0.00246
Haiku 4.5	$0.00000	$0.00123

Measured yesterday against content hash ca291c0d6787, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

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 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.

skill/commands/improve.md · 56 lines

How it starts

The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/humanise improve

Run one full cycle of the self-improvement loop: benchmark the skill's output, run the blind voice test, mine the accumulated evidence for weaknesses, and gate any proposed engine change. The loop's rules live in evals/self-harness-loop.md; this command executes them. Every stage is skippable, but say which stage was skipped and why in the final report.

Artefacts for the run land in profile/learning/runs/<UTC timestamp>/ (the soul; gitignored, never ships). Flags: --baseline also generates untreated comparison drafts; --skip-benchmark, --skip-indist to run a partial cycle.

a. Preflight

Held-in must be green before anything: python3 evals/assertions/selftest.py. Red means fix the engine first; an improvement cycle on a broken engine mines noise.
python3 scripts/build_voiceprint.py --status --json. No profile: offer /humanise init and stop. Stale voiceprint: rebuild it (humanise voiceprint --build). Note which channels are eligible for the pairwise test.
Create the run directory: profile/learning/runs/<timestamp>/.

b. Benchmark (rule compliance, with_skill vs baseline)

For each fixture in evals/evals.json, spawn an agents/eval-generator.md subagent in mode: skill, passing only the fixture's prompt, channel, audience_tag, and medium (never the assertions), with output path <run>/with_skill/eval-<id>.md. With --baseline, spawn a second generator per fixture in mode: baseline writing to <run>/baseline/eval-<id>.md.

Grade each directory: python3 evals/assertions/run_all.py --outputs <dir>. Read the exit codes correctly: baseline red is expected (the baseline exists to lose); with_skill hard failures are not a stop, they are mining signal for stage d. Summarise the delta (pass rates, which channels leak). Note in the report that an in-agent baseline is directional rather than a controlled experiment, and never gates anything.

c. Indistinguishability (the real voice gate)

For each eligible channel from preflight (at least 2 usable samples):

Read the full file on GitHub · 56 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.

yesterday First seen · 56 lines · 0 tokens per session scan A ca291c0d6787

Subscribe to this mod's changes

improve is a command published in the GitHub repository Nisus74/humanise (1 stars, last pushed 5d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,231 tokens. 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.

Testing Writing and docs agent-skills ai-tools ai-writing claude-code codex cursor

improve

Getting it into your agent

What it costs to keep this loaded

Grade A, and why

Nothing flagged

How it starts

/humanise improve

a. Preflight

b. Benchmark (rule compliance, with_skill vs baseline)

c. Indistinguishability (the real voice gate)

What this file has done since we first saw it

Other commands, from other repositories

test-e2e

test-integration

export

newpkg

tray-referencia-api

remove-bg