Resume Evidence Workflow

Resume Evidence Workflow

Run only the stage the user currently needs. Resume from existing files instead of restarting completed work. The full path is:

1. inventory existing material;
2. deep-dive and maintain a portable Markdown evidence bank plus reusable base-resume files;
3. decode the job description (JD) and map requirements to evidence;
4. select the best available resume skill;
5. rewrite, verify, and deliver an application-ready resume;
6. write newly confirmed or corrected facts back to the evidence bank.

When no job description (JD) is available, complete the evidence stage and preserve it for later tailoring. Do not imply that the complete product ends there.

Preserve evidence and user decisions

The evidence bank preserves what the user supplied and confirmed. The application resume may use a presentation choice explicitly requested by the user without rewriting the underlying evidence record.

Do not proactively add information that the user neither supplied nor requested. When the user explicitly asks for a different presentation of dates, titles, responsibilities, results, metrics, or other content, treat that request as the user's decision and pass it through to the selected resume skill. If a choice has a material downstream explanation risk, mention that risk once and continue without moralizing or requiring another confirmation. Read evidence and user agency when the evidence record and resume presentation differ.

Never place a real user's identifying facts in public documentation, demonstrations, screenshots, tests, or case libraries. Use wholly fictional identities, organizations, dates, metrics, and stories.

Stage A: inventory and deep dive

Inspect supplied resumes, profiles, portfolios, notes, transcripts, and existing evidence files before asking questions. Accept document-led and narrative-led entry. If speech or transcription is involved, read voice-input.md.

When a resume is supplied for the first time, read resume-workspace.md and offer to preserve a reusable 基础简历.md and 简历版式档案.md beside the evidence bank. After these files exist, treat them as the default resume source for later job-description (JD) tailoring. Do not ask the user to upload or reparse the same PDF for every job description (JD). Re-import only when the user says the source resume changed, requests a different design, or the saved files are missing or materially incomplete.

At the start of an experience, give a non-binding estimate of the likely core questions and name the main directions. Let one complete answer resolve several themes. Use clear natural language rather than resume or interview jargon; preserve professional terms the user already uses accurately.

Use three passes:

1. Recall: establish what the work was, why it began, what the user personally did, and what followed.
2. Lateral scan: read career-value-radar.md and select only plausible hidden branches such as monetization, governance, partnerships, systems, user insight, or reuse.
3. Selective deepening: read question-and-stop-rules.md. Continue only when an answer may reveal distinct value, clarify the evidence record, or change later resume selection.

After a substantial answer, briefly show only the neutral information that would be added or changed in the evidence bank, using slightly professional but not fully optimized resume language. This lets the user correct the record without rereading the file. Stop when added detail would be repetitive, trivial, unrecoverable, or unlikely to affect future use.

When creating or changing the bank, read evidence-schema.md. For substantial multi-project work, also read project-logic-and-interview-evidence.md. Use behavioral-question-coverage.md only as a lightweight event-discovery lens when process evidence is likely; do not turn this workflow into interview preparation.

Stage B: decode the job description (JD)

Keep the job description (JD) visible through the rewrite. Extract its business objective, must-haves, preferred signals, role family, seniority, market, language, ATS terms, and likely evidence standard.

Before requesting a resume, look for the reusable files defined in resume-workspace.md. Use 基础简历.md for current selection and wording, 简历版式档案.md for layout, and the evidence bank for deeper or newly relevant facts. A PDF is a calibration source, not a recurring prerequisite.

Before rewriting, show a requirement-to-evidence matrix distinguishing direct, adjacent, weak, and absent evidence. Explain intended emphasis, compression, movement, and gaps. Do not rewrite first and rationalize the match afterward.

Stage C: choose a resume skill

Read jd-and-skill-routing.md. Use installed skill descriptions to shortlist plausible candidates, then read only the complete SKILL.md files of the best two or three. Select one primary content-rewrite skill and, only if needed, one separate artifact/rendering skill.

If no suitable skill is installed, offer either a conservative direct rewrite using application-ready-resume.md, or a read-only search for up to three relevant public skills. Do not install or execute third-party code without the user's choice and required permission.