| name | resume-review |
| description | Review and improve resume content against the candidate's documented evidence and target role. Use for broad or focused resume critique, job-description tailoring, bullet refinement, before-and-after content comparison, seniority or narrative assessment, and claim-grounded interview preparation. Use resume-parsability instead for PDF extraction, ATS mechanics, Greenhouse parsing, reading order, fonts, tags, links, or layout experiments. Never invent facts, simulate independent reviewer consensus, or predict callbacks. |
Resume Review
Evaluate and improve resume content without turning judgment into fake
measurement. Trace claims to the repository's factual sources, separate fact
from inference, and keep the final recommendation proportionate to the
evidence.
Companion skills
- Use
resume-parsability when the request includes extraction, external ATS
behavior, PDF structure, field association, reading order, links, fonts, or
layout safety.
- Before changing Typst source, locate and follow the current
typst-author
skill.
- When visual layout matters, locate and follow the current
pdf skill and
inspect the rendered artifact.
Source precedence
Use evidence in this order:
- The user's requested outcome and supplied job description.
will_cygan_resume-data.json and the current rendered
will_cygan_resume.pdf.
work-experience/*.md and work-experience/ACRONYMS.md.
work-experience/RELEVANCE.md as the author's prioritization prior, not as
an objective score.
- Current primary or vendor documentation for claims about a named product or
process.
- Identified professional guidance or practitioner experience.
Do not use a job description, advice file, model completion, or relevance
score as evidence that the candidate performed work.
Reference router
Read the evidence model and project workflow for every task, then only the
mode-specific references needed.
Entry contract
Before analysis, establish:
Mode: review | tailor | compare | bullet | interview
Authoritative content and rendered artifact:
Target role or audience, if supplied:
Allowed edits: none | content | layout
Relevant supporting evidence:
Requested output and completion bar:
Infer these from the request and checkout when safe. Reviewing, explaining,
or diagnosing is read-only; edit only when the user asks for changes.
Core workflow
- Read the authoritative content and the relevant supporting entries.
- If visual hierarchy or density matters, inspect the current PDF rather than
inferring layout from JSON or Typst.
- Build a small evidence ledger before reaching conclusions.
- Apply only the review lenses relevant to the requested outcome. Do not
impersonate multiple reviewers or vote across personas.
- Prioritize findings by consequence, confidence, and actionability. Prefer
five strong findings to an exhaustive list of minor opinions.
- If editing, change the content authority rather than generated artifacts,
compile, inspect the final PDF, and run the checks required by the project
workflow.
- Report facts, inferences, heuristics, preferences, and unknowns distinctly.
Non-negotiable rules
- Never create or strengthen a claim beyond its documented evidence.
- Preserve actual employer titles. Describe target-role alignment through
truthful accomplishments and profile language, not title substitution.
- Quantify only when a reviewed source supports the number and its meaning.
- Do not manufacture keyword density, ATS scores, hiring probabilities,
recruiter scan times, or causal claims about callbacks.
- Do not report same-model personas as independent evidence or consensus.
- Treat missing support as unknown, partial, or gap; do not fill it by
inference.
- Preserve useful disagreement as a tradeoff when evidence does not settle a
judgment call.
Completion report
Lead with the recommended outcome. For each material finding include the
artifact location, evidence or counterevidence, confidence, and action. State
which checks were performed, whether the PDF was actually inspected, and what
remains untested.