Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/petar-djukic/writing-skillsnpx agentmods add skills/petar-djukic/writing-skills/tune-anchorsWrote 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/petar-djukic/writing-skills/tune-anchors)<a href="https://agentmods.dev/skills/petar-djukic/writing-skills/tune-anchors"><img src="https://agentmods.dev/badge/skills/petar-djukic/writing-skills/tune-anchors/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/petar-djukic/writing-skills/tune-anchors"><img src="https://agentmods.dev/badge/skills/petar-djukic/writing-skills/tune-anchors.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 | $0.00084 | $0.01801 |
| Opus 5 | $0.00042 | $0.00901 |
| Sonnet 5 | $0.00017 | $0.00360 |
| Haiku 4.5 | $0.00008 | $0.00180 |
Grade A, and why
tune-anchors 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 12d 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 — 181 lines — stays where its author put it; the contents beside it link to each section on GitHub.
tune-anchors
The objective
Given a writing-voice/ corpus and one or more target articles, answer:
which anchor-selection rule produces rewrites closest to the author's own
register?
A rule is an arm — a set of filters over the manifest: role=venue-voice,
tags~clipped, pre_ai=true. Different arms produce different anchor sets,
and the model copies the register of whatever it sees. The right arm is the
one whose anchors give the model a register worth copying.
The pipeline
Each full trial runs two model passes before measurement:
- match-voice (
drive.py) — voice rewrite using the arm's anchors - tighten-style (
tighten.py) — remove AI-register artifacts (passive stacks, nominalizations, filler) using the same model family shown transformation pairs, not rules - measure — register markers on the tightened output
The ranking reflects the final output quality, not the raw voice draft. Tightening is what removes the AI sound; ranking without it would penalize arms whose raw drafts carry fixable markers and reward arms whose markers survive tightening unchanged.
--no-tighten skips step 2 and measures the raw voice draft instead. Use it
to isolate the voice effect or to compare the tighten delta across arms.
When to run
- Writing-voice onboarding. After the manifest exists and before the first
real rewrite. Output is the
--anchor-tagsquery to use thereafter. - After the corpus grows. Pool sizes change, and guidance hardcoded to a count goes stale. The source study's recommendation (idea-factory#355) went stale exactly this way.
- After adding or removing tags. A tag query that was inert on the old corpus may now select a meaningful subset.
Prerequisites
| What | Required for | Notes |
|---|---|---|
writing-voice/ with manifest |
all commands | the corpus being tuned |
| Ollama with the target model | sweep (full) |
--dry-run needs no model |
Pangram key in .secrets/ |
verify only |
optional; verify is the last step |
What ships with it
7 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.
- scripts/ledger.py 7.6 KB runs code
- scripts/testdata/test_ledger.py 6.7 KB runs code
- scripts/testdata/test_preserve_drafts.py 3.5 KB runs code
- scripts/testdata/test_rank_verify.py 11 KB runs code
- scripts/testdata/test_sweep.py 11 KB runs code
- scripts/testdata/test_tune_anchors.py 6.5 KB runs code
- scripts/tune_anchors.py 20 KB runs code
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.
- 12d ago First seen · 181 lines · 84 tokens per session scan A fe9180e2aba4
tune-anchors is a skill published in the GitHub repository petar-djukic/writing-skills (4 stars, last pushed 8d ago), licensed MIT. It adds 84 tokens to every session and 1,801 once invoked, about $0.0004 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
humanizer
Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague…
structural-humanizer
Remove the discourse-level (structural) signs of AI writing that survive surface editing: stated lessons and moral-of-the-story closers, tidy single-track arcs, embodied-emotion performance ("chest tightened"), vague allusions instead of named references, unbroken linear structure, and shape convergence across pieces.…
story-import
A tool for turning an existing novel into a structured writing project. It analyzes the book and organizes its characters, settings, plot plans, chapters, and tracking data for later writing.
story-review
A review workflow for finding story problems from several viewpoints, including issues with structure, characters, wording, and fictional world rules. It can use multiple reviewer agents or work alone.
story-long-analyze
A long-form fiction analysis workflow for breaking down a novel’s opening chapters, characters, pacing, turning points, relationships, and overall structure.
story-deslop
A writing editor for Chinese web novels that detects writing patterns often associated with AI-generated text and makes the prose feel more natural.