Registers a candidate's CV (Markdown or plain text) and extracts individually-citable claims from it. Every extracted claim is stored with confidence 'user_claim' — it is what the candidate says about themselves, not independently verified.

Indexes a public (or token-accessible) GitHub repository: recent commits, pull requests, README/architecture docs, and dependency manifests. Everything pulled directly from GitHub is stored as 'verified' evidence because it is directly observed, not inferred.

Manually attaches an evidence item to an indexed project (e.g. an architecture decision explained in conversation). Defaults to 'inferred' confidence since it wasn't pulled directly from a verifiable source.

Free-text search across all stored evidence (commits, pull requests, docs, dependency manifests, CV claims and manual notes). Every result carries its confidence tag (verified / inferred / user_claim / missing) so the caller knows how much to trust it.

Stores a pasted job description and extracts structured requirements (skills, tools, competencies) using bullet parsing and a curated keyword dictionary. No LLM is used here — the connected client can add further reasoning on top of these structured requirements.

Scores every requirement extracted for a job against all stored evidence (indexed repositories and/or CV claims), returning a match strength (strong / moderate / weak / unproven) plus the specific evidence cited for each requirement.

Re-runs requirement matching for a job and returns only the requirements with weak or unproven support, each with a concrete suggestion for what evidence to add.

Builds a Situation/Task/Action/Result outline for a competency and project, scaffolded entirely from stored evidence (commits, PRs, docs). Never invents narrative detail: gaps are listed explicitly in 'missing_information' rather than filled in. The connected client should turn this outline into flowing prose.

Generates likely interview questions from a job's requirements, prioritising requirements with the weakest evidence so preparation time is spent where it matters most.

Rule-based (non-LLM) check of a candidate's practice answer: STAR structure presence, first-person action verbs, measurable results, length vs. a target word count, and keyword relevance to the target competency. Returns a score and specific, actionable feedback rather than a black-box grade.

Assembles the complete evidence-backed preparation pack for a job: requirement matches, evidence gaps, and generated interview questions. Returns either JSON or a Markdown report.