Agent skill · Audited from source
Style Harmonizer
skill · WILLOSCAR/research-units-pipeline-skillsLicence not declaredAudited 19 Oct 2026 · source 81994516a430
Is it safe? Yes — SAFE, grade B (89/100). We read this skill's source at commit 81994516a430 and found no filesystem, network or shell access and no hard-coded secrets. What it does: De-slot and harmonize paper voice across `sections/*.md` without changing meaning or citation keys.
Cite
Cite this page
ModelsAtlas. “Style Harmonizer (WILLOSCAR/research-units-pipeline-skills)”. Retrieved 19 Oct 2026. https://modelsatlas.com/skills/willoscar/style-harmonizer
Trust audit
Audit 0.4.1 · source read 19 Oct 2026
SAFE means nothing in the source contradicts what the skill says it does. Grade B is shared by 16,667 of 17,602 audited skills.
What the source does
- FilesystemNone observedNo reads or writes in the source
- NetworkNone observedNo outbound calls in the source
- ShellNone observedNo commands run in the source
- DependenciesPinnedNothing floats
- SecretsNone foundNo keys in source
Five audit layers
| Layer | What we check | Result | Findings |
|---|---|---|---|
L0 Provenance | Repo owner, licence, archive status, activity | WARN | 2 |
L1 Static source | Secrets, path traversal, supply chain, hidden code | PASS | 0 |
L2 Instructions | Prompt injection, authority framing, hidden directives | PASS | 0 |
L3 Inventory | The files and declarations match the source | PASS | 0 |
L4 Sandbox | Run in isolation and watch behaviour | SKIPPED why | 0 |
Gates applied: no behavioural pass, no license.
Findings (2)
MEDIUMInventory / provenanceinv.symlink · .claude/skills
.claude/skills
- Why
- link not followed
- Fix
- —
LOWInventory / provenanceinv.no_license
- Why
- no LICENSE file and no repo licence
- Fix
- add a licence
Audit 0.4.1 · source sha 81994516a430 · 19 Oct 2026Report an issue or request a re-scan
Provenance
Who published it, and what the verdict is pinned to
- PublisherWILLOSCAR GitHub userOwner of the repository this skill is read from
- Source folderWILLOSCAR/research-units-pipeline-skills/.codex/skills/style-harmonizerThe exact folder we audited
- LicenceNot declaredAs the repository states it
- Commit pinned for this verdict81994516a430A new audit runs when the source changes
- Repository statusNot archivedLast push 14 days ago
- Sandboxed runNot runSkills are read, not executed; layer L4 is skipped
What it tells the agent
Full source on GitHub ↗Its description: De-slot and harmonize paper voice across `sections/*.md` without changing meaning or citation keys.
---
name: style-harmonizer
description: "De-slot and harmonize paper voice across `sections/*.md` without changing meaning or citation keys."
---
# Style Harmonizer (de-slot editor)
## Triggers & routing
- **Trigger**: style harmonizer, de-template stems, remove slot phrases, discourse stems, 写作风格统一, 去槽位句式, 去生成器味.
- **Use when**: `writer-selfloop` is PASS but `output/WRITER_SELFLOOP_TODO.md` flags Style Smells (e.g., repeated count-based openers), or the draft reads like many sections share the same rhythm.
Purpose: remove subtle generator-voice signals that can survive structural gates.
This skill is not a full rewrite. It is a targeted rewrite queue:
- only touch the specific `sections/*.md` files flagged under `## Style Smells`
- keep facts and citation keys unchanged
## Inputs
Required:
- `output/WRITER_SELFLOOP_TODO.md` (Style Smells section)
- the referenced `sections/*.md` files
Optional (helps you stay in-scope while rewriting):
- `outline/writer_context_packs.jsonl` (allowed citations + opener_mode hints)
## Output
Run this targeted pass after `writer-selfloop` and before `opener-variator` and
logic polish. If major edits happened since the report, rerun
`writer-selfloop` so `## Style Smells` reflects current text. The deterministic
script is a certification adapter: it refuses to create the marker while the
latest report still names flagged files or predates any `sections/*.md` file.
Perform the semantic rewrite through this Skill or the responsible upstream
writer, rerun `writer-selfloop`, then retry the adapter. A passing marker records
the certified Section-tree SHA256.
- Updated `sections/*.md` files (same filenames; still body-only; no headings)
- Re-running `writer-selfloop` is the audit trail (Style Smells should shrink).
- Create `sections/style_harmonized.refined.ok` (empty file) when you are done (pipeline contract signal; required if this unit is marked DONE).
## Role cards (use explicitly)
### Style Harmonizer (editor)
Mission: remove slot phrases and stem repetition while keeping meaning unchanged.
Do:
- rewrite the surface form (opener/closer/cadence), not the claim
- keep each paragraph content-bearing (argument bridge, not navigation)
- prefer small local edits over global style refactors
Avoid:
- adding new factual claims or new citations
- moving citations to different paragraphs or different subsections
- rewriting a thin section instead of routing upstream for more evidence
### Evidence Steward (skeptic)
Mission: prevent style work from becoming content drift.
Do:
- after each rewrite, spot-check that every cited claim still matches the same sentence
- if you feel forced to add new material to make prose sound better, stop and route upstream
## Common style smells and how to fix them
### 1) Count-based opener slots (Two limitations..., Three takeaways...)
Why it is high-signal: it creates a reusable sentence slot that repeats across H3s.
Rewrite moves (choose one):
- Integrate the caveat into a contrast paragraph (last sentence): state the boundary that changes interpretation.
- Use a single caveat sentence opener (no counting), but rotate across H3s (avoid repeating the same stem): "These results hinge on ..." / "Interpretation depends on ..." / "Evidence is thin when ..." / "A caveat is that ..." (use sparingly).
- If enumeration is truly needed, hide the count: use two coordinated clauses in one sentence, or vary the syntax (do not repeat across sections).
Mini example (paraphrase only):
- Bad: `Two limitations stand out. First, ...`
- Better: `These results hinge on ...; this matters because it changes how results transfer across protocols.`
### 2) Reused discourse stems (The key point is that ...)
Rewrite moves:
- Replace with one of: "A practical implication is that ...", "One takeaway is that ...", "A useful way to read these results is ...".
- Change cadence: split into a short claim sentence plus a follow-up sentence with the condition/why.
### 3) Same opener cadence across many H3s
Rewrite moves:
- Switch opener mode for the section (tension-first / decision-first / protocol-first / contrast-first).
- Replace generic connectors (Additionally/Moreover) with content-bearing pivots ("At the protocol level, ...", "Under budget constraints, ...").
### 4) Overview / narration openers ("This section provides an overview ...")
Why it is high-signal: it reads like a generated ToC narration rather than a paper argument.
Rewrite moves:
- Replace "overview" narration with a content-bearing lens: tension/decision/failure/protocol/contrast.
- Keep the first sentence falsifiable: name the constraint and why it matters (not what you are about to do).
Mini example (paraphrase only):
- Bad: `This section provides an overview of tool interfaces for agents.`
- Better: `Tool interfaces define what actions are executable; interface contracts therefore determine which evaluation claims transfer across environments.`
### 5) Paragraphs repeatedly starting with connector adverbs (Moreover, In addition, Therefore, Overall, As a result)
Why it is high-signal: the prose starts to sound mechanically stitched (each paragraph begins with the same connective), even when the content is solid.
Rewrite moves:
- Keep the logical relation, but move the connector into the sentence: start with the subject (e.g., "Tool catalogs also ..."), then add the relation mid-sentence ("..., which in turn ...").
- Use clause shapes instead of adverb openers: "While ... , ..." / "Although ... , ..." / "Because ... , ...".
- When summarizing, avoid "Overall," as a default label; state the conclusion directly as a claim sentence.
Mini example (paraphrase only):
- High-signal smell: "Moreover, ..." (repeated across multiple paragraphs)
- Better: start with the content noun phrase ("One implication is ..." / "A practical constraint is ..."), then express the relation inside the sentence.
### 6) Internal shorthand leaking into paper voice ("token(s)" used as a protoExcerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.
Install
Pinned to the audited commit 8199451
git clone https://github.com/WILLOSCAR/research-units-pipeline-skills git -C research-units-pipeline-skills checkout 81994516a430
The author documents no install command, so this uses the repository directly. Our verdict covers commit 81994516a430 only.
Works in which agents?
What its own text says, not a test
| Agent | Status | More |
|---|---|---|
| Codex | Mentioned in its text, not tested | Skills for Codex · About Codex |
Audit history
Every audit, every source change
- ✓SAFE B 89 / 100
first audit · source
81994516a430
README badge
For maintainers · shows the latest audit
[](https://modelsatlas.com/skills/willoscar/style-harmonizer/)
The badge shows our verdict and grade for this skill, and links back to this page. It changes when the audit does.
Similar skills
All automation skills →WILLOSCAR/research-units-pipeline-skills
Compile the Beamer tutorial deck and write a build report.
WILLOSCAR/research-units-pipeline-skills
Generate a Beamer slide deck from the final tutorial and approved module structure.
WILLOSCAR/research-units-pipeline-skills
Build a retrieval-informed chapter skeleton (`outline/chapter_skeleton.yml`) from taxonomy/core scope before stable H3 decomposition.
WILLOSCAR/research-units-pipeline-skills
Score an existing reader-facing deliverable and emit a deterministic PASS/FAIL scorecard plus bounded repair plan; use after synthesis, not to rewrite content.
WILLOSCAR/research-units-pipeline-skills
Review and record one pending human checkpoint in a research Workspace; use when a HUMAN Unit is blocked on an `Approve C*` decision, and never treat silence or artifact existence as approval.
WILLOSCAR/research-units-pipeline-skills
Lock an ideation run into a single-source-of-truth brainstorm brief (`output/trace/IDEA_BRIEF.md`) and a replayable multi-query plan (`queries.md`).
Add this badge to your README
[](https://modelsatlas.com/skills/willoscar/style-harmonizer/)
<a href="https://modelsatlas.com/skills/willoscar/style-harmonizer/"><img src="https://modelsatlas.com/skills/willoscar/style-harmonizer/badge.svg" alt="ModelsAtlas: SAFE B"></a>
Questions about Style Harmonizer
What does the Style Harmonizer skill do?
Is Style Harmonizer safe to install?
What can Style Harmonizer access on my machine?
Which assistants does Style Harmonizer work with?
How current is this page?
Sources
- github.com/WILLOSCAR/research-units-pipeline-skills/tree/main/.c — the audited source at commit 81994516a430 and every finding aboveObserved · 19 Oct 2026
Spotted something wrong? Report a correction · Methodology
Keep going in your AI
Opens with this page as the source. We add nothing else to the prompt.
- What does the Style Harmonizer skill do?
- Is Style Harmonizer safe to install?
- What can Style Harmonizer access on my machine?