Documents

Queue text extraction for your own member upload. Requires a verified tenant_auth session (wst_ Bearer). The file must be a gated-files member upload by you, in your tenant. Idempotent: offering the same file again returns the same extraction. Returns { extraction_id, status, record_id, existing }.

Status and result of one of your own extractions. Requires a verified tenant_auth session. Returns status (queued | extracting | ok | partial | failed), method, quality, errors, notes and per-page metadata; the text only with include_text=true.

Skill candidates from one of your own extractions, for review. Requires a verified tenant_auth session. Returns candidates_status (queued | running | ok | failed; null before the first run) and the candidates: occupations, software and skills, each grounded in O*NET with the CV lines as evidence. Default filter: pending. After a new candidate run, undecided candidates get new candidate_ids — fetch the list again before deciding. If the project has the AI review on, the list also carries ai_status (queued | running | ok | skipped | failed; null when off) and a candidate can carry extra.ai { v, verdict: keep | drop, reason, quote }: a marking with a quote from your own resume, not a decision — you decide.

Confirm or reject skill candidates from one of your own extractions. Requires a verified tenant_auth session and skill confirmation enabled for the project. A confirmed candidate becomes a record in the configured skill entity under your identity, with its computation trace; a rejected candidate stays out of your profile. A decision is final: deciding again returns skipped, an unknown or outdated candidate_id returns not_found. Returns { extraction_id, results, counts }.

Retract skills you confirmed earlier. Requires a verified tenant_auth session and skill review enabled for the project. The skill record itself is kept with its computation trace; a retract event is added under your identity and the skill no longer counts in your profile. A skill that is not yours or does not exist returns not_found; a skill you already retracted returns skipped. Returns { results, counts }.

Search O*NET for skills, software or occupations to add to your own profile. Requires a verified tenant_auth session and skill review enabled for the project. Returns up to 10 options per kind, each { kind, ref_id, label, category, match_type, match_score }, plus the O*NET release and the attribution that must be shown with the results. A search that finds nothing is counted anonymously, without any link to you, as a signal for improving the matching.

Add skills, software or occupations your resume did not show, chosen from search-skills. Requires a verified tenant_auth session and skill review enabled for the project. Only O*NET references are accepted: the server looks each one up again and stores the O*NET label, never text from the request. Stored under your identity, marked as added by you (match_type member_added). A reference you already have in your profile returns skipped; an unknown reference returns invalid_ref. Returns { results, counts }.

Your own effective skill profile: the skills you confirmed from your resume and the ones you added yourself, without the skills you retracted. Requires a verified tenant_auth session and skill confirmation enabled for the project. Each skill comes with { resume_skill_id, kind, label, category, ref_id, source: resume | added, evidence, stored_at }; evidence holds the resume lines it came from. With include_trace=true each skill also carries its computation trace (match type and score, method version, O*NET release, candidate, extraction and resume record). Returns { skills, total, counts, retracted_count, truncated, include_trace, release, attribution }; the attribution must be shown with the results.

Project settings. ingest_enabled switches member ingest on (default off). record_entity names the MAPI entity for the per-upload anchor record; it must have the fields gated_file_id, extraction_id and original_filename, an owner_field and create: { verified: self }. skill_entity names the MAPI entity for confirmed candidates; it must have the trace fields item_kind, onet_ref, onet_category, match_type, match_score, evidence, candidate_id, extraction_id, method_version and onet_release, an owner_field and create: { verified: self }. skill_field_map maps anchor_record_id, label and rank to field names of that entity. skill_event_entity names the MAPI entity for append-only skill events (retract); it must have the fields resume_skill_id, action, reason, candidate_id and extraction_id, an owner_field, create: { verified: self }, list: { verified: own } and no member update. All checked on save. Send an empty record_entity to switch the anchor record off, an empty skill_entity to switch confirmation off, an empty skill_event_entity to switch retract off. ai_enabled switches the AI review of skill candidates on (default off): after each candidate run the resume text, with name and contact details removed, goes with the candidates to the project's own openai integration (vault key), and each pending candidate gets a keep or drop marking in extra.ai. ai_model picks the model (default gpt-5.4-mini); ai_daily_cap limits model calls per project per day (default 100). Needs the AI database columns; without them these keys return 409. The response carries ai { available, enabled, model, daily_cap, effective }.

Full extraction for the owner: status, text, per-page text and methods, quality, engines, errors and notes. For review and for the skill-extraction step. Also the candidate run (candidates_*) and the AI review: ai_status, ai_method, ai_at and ai_trace (model, versions, tokens, duration, counts; no resume text).

Which extraction programs are installed on the serving node, with versions and OCR languages. Diagnostics after an install or deploy.

No input parameters required.

Skill candidates found in one extraction: occupations (from job titles), software and skills, each grounded in the occupations reference, with the CV lines as evidence, the method version and the O*NET release. Candidates are not profile data until the member confirms them. With the AI review on: ai_status per extraction and extra.ai per candidate, as in my-candidates.

Queue a new candidate run for an extraction, for example after the method improved. Candidates the member already confirmed or rejected stay as they are. With the AI review on, the new candidates are reviewed again (one model call).