resume-base-selector
Skill0 starsChanged 52 days agoMITView on GitHub
Match a complete job description to one or more master resume PDFs/DOCX files, compare job-family fit, and select the strongest base before tailoring. Useful for data, BI, product analytics and career-change applications. With one resume, assess fit without inventing alternatives.
In this repositoryOne other file it gives its agents.
- resume-tailorSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 7 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 28
Import candidates with resume-workflow/scripts/import_resumes.py and validate source-bound evidence following Step 1b of …
- Distinctive: line 37
Treat attached resumes as comparison inputs only.
- Very specific: line 39
Keep originals unchanged at their local paths.
- Distinctive: line 46
Read references/evaluation-rubric.md before scoring.
- Distinctive: line 51
Prefer role-relevant accomplishments and scope over keyword repetition, layout, or prose polish.
- Distinctive: line 52
Treat scores as relative evidence-fit judgments, never an ATS score or hiring probability.
- Distinctive: line 57
Do not invent experience, merge contradictory facts, upgrade seniority, or claim that a requirement is met when the resumes do not show it.
- Distinctive: line 70
Use explicit language such as "evidence not found" for gaps.
- Very specific: line 74
After every concrete selection with a complete JD, create a resume-selector-handoff/v1 object using references/handoff-contract.md and the …
- Very specific: line 81
Use the lowercase hyphenated slug from prepare and write applications/<slug>/handoff.json.
- Distinctive: line 83
Never overwrite a different existing handoff.
- Very specific: line 93
Do not start tailoring in this skill.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Resume Base Selector
Choose the master resume that already contains the strongest truthful evidence for the target role and will require the least structural rewriting. Selection is the default deliverable; do not tailor the resume unless the user also asks for tailoring (that is the resume-tailor skill).
Resolve the inputs
- Obtain the complete job description. Capture the title, company, core outcomes, responsibilities, minimum qualifications, preferred qualifications, seniority signals, domain, and location or work-authorization constraints when stated.
- A saved file is the reliable path:
.txt,.md,.html, or.json. A URL is best effort only. No signed-in browser session is assumed. Run:
python3 resume-workflow/scripts/capture_job_description.py --url <url> --out <path>/job-description.txt- If the URL returns a login wall, a partial page, an expired posting, or anything shorter than the configured minimum, the capture fails on purpose. Ask the user for the complete JD text or a saved page. Never reconstruct requirements from a title, a snippet, or your own knowledge of the company.
- A saved file is the reliable path:
Open the application bundle with the deterministic runner, which also stores the JD and refuses to silently reuse a directory that belongs to a different job:
python3 resume-workflow/scripts/run_resume_workflow.py prepare \ --company "<Company>" --role "<Role>" --jd-file <path-to-complete-jd>- Import candidates with
resume-workflow/scripts/import_resumes.pyand validate source-bound evidence following Step 1b of.github/agents/resume-application.agent.md. Users never need to write JSON. Build the candidate set from:- resume files the user identifies for this comparison;
- resumes attached in the current conversation;
- a library the current user explicitly asks you to use.
- Exclude cover letters, historical one-off applications, duplicate exports, and unreadable files. With one usable resume, assess its fit, select it, and report "single candidate; no alternatives compared." Never invent a second version.
- Treat attached resumes as comparison inputs only. Add or replace a persistent library entry only when the user explicitly asks to save, register, or update a master resume. Never overwrite an original resume.
Keep originals unchanged at their local paths. Use import_resumes.py import for
PDF/DOCX/TXT, then transcribe and validate evidence as the agent instructions specify.
The legacy layout-specific extractor is not the public onboarding path. An optional
registry may reference imported evidence; no registration/configuration is required.
Compare the job and resumes
Read references/evaluation-rubric.md before scoring.
- Normalize each resume into job family, likely level, functional strengths, domain exposure, leadership scope, tools or methods, and quantified outcomes.
- Separate hard gates from preferences. A stated must-have with no supporting evidence outweighs several matching keywords.
- Score evidence that is actually present. Absence means "not evidenced in this resume", not that the person lacks the qualification.
- Prefer role-relevant accomplishments and scope over keyword repetition, layout, or prose polish.
- Treat scores as relative evidence-fit judgments, never an ATS score or hiring probability.
- Estimate tailoring effort. Penalize a base that would need a new professional identity, reordered career narrative, or extensive bullet replacement.
- Compare across the entire resume, including summary, skills, experience, and selected projects. Cite the section or role that supports each important judgment.
- Still choose one primary base. If another resume contains useful verified material, identify it as a secondary donor rather than returning an indecisive tie.
Do not invent experience, merge contradictory facts, upgrade seniority, or claim that a requirement is met when the resumes do not show it.
Report the decision
Lead with the recommendation. Then provide:
- the selected resume filename and its job family;
- a compact comparison table showing total score, strongest matches, largest gaps, and expected tailoring effort for every candidate;
- the decisive reasons, tied to specific resume evidence and job requirements;
- missing or weakly evidenced must-haves that remain risks regardless of the base;
- any secondary resume worth mining for specific bullets or skills;
- confidence (
high,medium, orlow) and what would change the decision.
Use explicit language such as "evidence not found" for gaps. If the result is close, apply the tie-breakers in the rubric and say that the choice is narrow. End by offering to tailor the selected base only after the selection is clear.
Produce the tailoring handoff
After every concrete selection with a complete JD, create a resume-selector-handoff/v1 object using references/handoff-contract.md and the canonical schema at ../../../resume-workflow/schemas/handoff-schema.json.
- Preserve the complete JD text and capture timestamp so the downstream workflow does not depend on a live page. The text must match
applications/<slug>/job-description.txt. - Resolve the selected base and donors through the resume library; include both the original resume path and
structured_path. - Carry forward normalized requirements, hard-gate status, the rubric's seven scoring dimensions, confidence, decisive reasons, and known gaps.
- Recommend
balancedby default; useatsorhmonly when requested or clearly justified. - Use the lowercase hyphenated slug from
prepareand writeapplications/<slug>/handoff.json. - For a batch, create one handoff per job plus a manifest recording job URL, slug, base id, confidence, and handoff path.
- Never overwrite a different existing handoff. Reuse one only for the same job, and regenerate it when the captured JD changes.
Validate before reporting that the handoff is ready:
python3 resume-workflow/scripts/run_resume_workflow.py validate-handoff --slug <slug>The runner checks the schema, the paths, the comparison/score consistency, and that the handoff JD matches the captured JD. Alternatively, validate a standalone file with scripts/validate_handoff.py. If the environment is read-only, return the complete JSON object in the response and state that it was not written to disk.
Do not start tailoring in this skill. Hand off to resume-tailor only after validate-handoff passes.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

