cv-witness
Agent skill for drafting and reviewing CVs from verified experience. Candidate-led wording, phrase-overlap review and local PDF tools.
Install / Use
npx skills add chindris-mihai-alexandru/cv-witnessInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
Content & MediaSupported Platforms
Our assessment of cv-witness
cv-witness scores 82/100 on our quality scale, 947th of 1,153 Content & Media skills we index.
Its SKILL.md is 16 KB long, well organised into 17 sections with 10 code examples: a thorough specification that gives an agent plenty to work with.
It has 2 GitHub stars, so there is little community track record yet; judge it on its content.
Maintenance, license and trust
- The repository was last updated today, so cv-witness is actively maintained.
- It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-11-15. Automated pattern scan on 2026-11-15. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
cv-witness compared with similar skills
All 4 of these similar skills score higher than cv-witness; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| cv-witness (this skill)by chindris-mihai-alexandru | 82 | 2 | today | SKILL.md |
| siyuanby siyuan-note | 100 | 47.3k | today | MCP Server |
| x402by browser-use | 100 | 121.0k | 3d ago | SKILL.md |
| algorithmic-artby anthropics | 100 | 177.9k | 55d ago | SKILL.md |
| pptxby anthropics | 100 | 177.9k | 55d ago | SKILL.md |
Frequently asked questions
- How do I install cv-witness?
- Run
npx skills add chindris-mihai-alexandru/cv-witness. The install tabs above show the steps for each supported agent. - Which AI agents does cv-witness work with?
- It is written for Claude Code, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is cv-witness safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is MIT-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is cv-witness still maintained?
- The repository was last updated today, so cv-witness is actively maintained.
Skill content
View source on GitHubAn open-source agent skill for drafting and reviewing a CV from your real experience. It helps you select relevant work, trace claims to a source and inspect the exported document. You review the facts and wording before applying.
How cv-witness differs
The job posting guides emphasis and ordering. It should not become a script for rewriting your history. Describe the work you did, keep accurate technical terms, and flag missing facts instead of inventing them.
The skill asks your agent to:
- Check dates, credentials, metrics and contribution claims against your sources. Verify a public pull request's merge state before calling it merged. Before delivery, run
scripts/check_cv_numbers.pyagainst the approved source files; this lexical check flags unsupported numbers but does not validate their context or meaning. - Before tailoring from a job URL, run
scripts/check_posting_liveness.py <url> --expect "Role Title". It returns live, dead or suspicious; verify suspicious or client-side-rendered postings in a browser rather than treating uncertainty as closure. - Optionally triage a role first: judge each posting requirement against your verified facts, then let
scripts/fit_score.pyverify the quoted evidence and return a fit band with an uncertainty range. See grounded fit scoring; it is not an ATS score. - Use
[FILL: ...]markers in a draft when information is missing, then resolve or remove them before delivery. - Run the phrase-overlap script and review shared wording in context. A common technical phrase is not proof of copying and need not be changed just to avoid a match.
- Inspect the exported text, page count, placeholders and layout before sending the CV.
These are workflow instructions, not guarantees an agent cannot bypass. The PDF renderer creates a file; it does not automatically certify its contents. No ATS parsing result, ranking, exact voice match or interview is promised.
Install
One root SKILL.md is shared by every installation. The template and scripts live beside it, so there are no duplicate skills or recursive directory links.
AdaL
git clone https://github.com/chindris-mihai-alexandru/cv-witness.git ~/.adal/skills/cv-witness
Claude Code
Send these as two separate commands inside Claude Code:
/plugin marketplace add chindris-mihai-alexandru/cv-witness
/plugin install cv-witness@cv-witness
The plugin manifest points to the root skill. Alternatively, install a personal skill:
git clone https://github.com/chindris-mihai-alexandru/cv-witness.git ~/.claude/skills/cv-witness
Codex
git clone https://github.com/chindris-mihai-alexandru/cv-witness.git ~/.agents/skills/cv-witness
For other hosts, follow their skill-discovery instructions or ask the agent to read the root SKILL.md. If the destination already contains a clone, update it instead of overwriting your private sources.
Before first use, review the personal configuration and authoring preferences in SKILL.md. Set your own CV-guide, application-log and output paths; the owner's Cabinet paths and Quick preference are examples, not requirements for your setup.
Compatibility and tools
Agent hosts
<p> <a href="https://docs.sylph.ai/"><img alt="AdaL skill" src="https://img.shields.io/badge/AdaL-skill-2b3542?style=flat&labelColor=1b222c"></a> <a href="https://code.claude.com/docs/en/plugins/overview"><img alt="Claude Code plugin" src="https://img.shields.io/badge/Claude%20Code-plugin-2b3542?style=flat&labelColor=1b222c&logo=anthropic&logoColor=white"></a> <a href="https://learn.chatgpt.com/docs/build-skills"><img alt="Codex skill" src="https://img.shields.io/badge/Codex-skill-2b3542?style=flat&labelColor=1b222c"></a> </p>AdaL, Claude Code and Codex have installation paths above. Other hosts may read the same skill format; they are not all independently tested.
Local tools
<p> <img alt="Python 3.11+" src="https://img.shields.io/badge/Python-3.11%2B-2b3542?style=flat&labelColor=1b222c&logo=python&logoColor=white"> <img alt="Chromium PDF rendering" src="https://img.shields.io/badge/Chromium-PDF-2b3542?style=flat&labelColor=1b222c&logo=googlechrome&logoColor=white"> <a href="https://cli.github.com/"><img alt="GitHub CLI" src="https://img.shields.io/badge/GitHub-CLI-2b3542?style=flat&labelColor=1b222c&logo=github&logoColor=white"></a> </p>render.sh delegates to the shared Python 3.11+ renderer, which discovers Chromium-family browsers on Linux and macOS. Poppler supplies pdftotext and pdfinfo for pre-delivery checks; pdftoppm creates sample previews. Authenticated GitHub CLI is used for contribution checks, not for rendering or phrase review.
Optional services
<p> <a href="https://github.com/rezi-io/rezi-mcp"><img alt="Rezi MCP optional" src="https://img.shields.io/badge/Rezi%20MCP-optional-2b3542?style=flat&labelColor=1b222c"></a> <img alt="Amazon Quick optional" src="https://img.shields.io/badge/Amazon%20Quick-optional-2b3542?style=flat&labelColor=1b222c"> </p>Rezi can provide resume facts from an account you choose. Writes require your explicit approval and a read-back check. Amazon Quick is a documented host-native DOCX workflow, not a bundled integration. Neither service is required for the local PDF path.
Usage
Run the command examples from your cloned skill directory. The Python tools use the standard library; PDF export also needs a Chromium-family browser and Poppler. Keep real inputs and outputs outside the public examples/ and assets/ folders.
Ask your agent:
"Draft a CV from my verified experience for this role: [paste the posting URL or text here]. Keep my voice and show me the wording before exporting."
Provide a master resume, the latest candidate-approved CV (or a specific design reference), and verification notes privately. The approved CV is the default design baseline across authoring tools; ask before changing its layout or section order. Review the selected facts and approve the draft before rendering. The agent workflow uses the output directory configured in SKILL.md; standalone build and render commands require an explicit output path.
For phrase review:
python3 scripts/jd_overlap.py CV.pdf posting.txt
The script compares literal word sequences. --strict fails on any shared run of four or more words, including ordinary industry phrases. Reports give counts and positions, including repeated occurrences; --show-phrases is an explicit opt-in for private wording. A pass is not an ATS or originality certification.
For checked output, use sh render.sh approved.html output.pdf a4 --posting posting.txt. The renderer validates a temporary PDF before publishing and preserves previous output if any check fails. It rejects external/resource-loading content in the HTML, including CSS escapes that disguise @import or url(). This narrow preflight is not a general-purpose hostile-HTML sandbox. Selecting a posting uses strict overlap checks by default; choose --overlap-policy review explicitly when shared generic wording should be reviewed rather than blocked.
For a PDF exported from another authoring tool:
python3 scripts/check_cv.py exported.pdf \
--paper a4 --max-pages 1 \
--expected-text "Alex Morgan" \
--ordered-text "Professional Summary" \
--ordered-text "Skills" \
--ordered-text "Professional Experience" \
--strict-ordered-text \
--posting posting.txt
Add critical text as --expected-text arguments and section headings in approved order as repeated --ordered-text flags. Add --strict-ordered-text for standalone heading lines in generated CVs. This checks extracted landmarks, not every ATS's reading order or field parsing. Compare the export visually with the approved CV and review any employer autofill before submission. Parsing varies by platform: Greenhouse documents reasons parsing can fail, while Rezi describes its score as guidance, not a particular ATS result. Exit codes are 0 for a policy pass, 1 for failed checks and 2 for input or tool errors. See delivery checks for limits and privacy-safe diagnostics.
Repeatable career-lane templates
For a local CV with more than the legacy template's fixed role slots, provide approved structured wording:
python3 scripts/build_cv.py approved-cv.json output.pdf \
--lane engineering --paper a4 --max-pages 1 \
--posting selected-posting.txt
Engineering, Operations/Support, Research, Management, Security and Customer-Success are section-order presets over one single-column layout. Roles repeat, empty optional sections are omitted, and A4/Letter exports use the same delivery checks. The builder checks that its selected section headings extract in the chosen order. These are product layouts, not replacements for a candidate's existing design; obtain approval before switching. The single-column preset is a parsing-risk recommendation, not a requirement to redesign an approved two-column CV. Longer CVs need an explicit approved page limit; text is never dropped or silently shrunk. See the template contract and fictional structured example.
Explore the repository
Open the interactive project guide for the animated workflow, searchable source map, three guided reading tours and eight fictional template PDFs.
<p align="center"> <a href="https://chindris-mihai-alexandru.github.io/cv-witness/assets/explore/workflow.html"><img src="assets/explore/workflow-preview.svg" alt="Approved wording passes through the shared layout, temporary PDF and delivery checks. Failure preserves prior output; passing output still needs visual and factual review." width="520"></a> </p>Workflow · Source map · Regeneration and limits
The README preview has a fixed light background and dark labels in either GitHub theme. Open it for the interactive workflow, whose theme and pause controls remain available. The full maps work offline and respect reduced motion. They describe verified workflow and public source structure, not live telemetry; private resumes and writing corpora are excluded.
Optional candidate voice reference
Use your own selected writing as a private style reference. The local corpus tool accepts text/Markdown (including Cabinet pages and Obsidian notes), Sanity exports, and a common original-text format for other CMS or memory exports. It filters authors and message roles, excludes labelled generated/quoted content, and measures writing habits without training a model or uploading text.
python3
Truncated for display — read the full file on GitHub.
Related Skills
siyuan
47.3kAn open-source, privacy-first, self-hosted knowledge workspace where humans and AI agents work together 开源、隐私优先、自托管的知识工作空间,让人与智能体在此协作
x402
121.0kSet up Browser Use Cloud payments with x402 — pay per request from a crypto wallet (USDC on Base mainnet), no signup or API key.
algorithmic-art
177.9kCreating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems.
pptx
177.9kUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an em…
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
