[New Skill]: office/application_auditor — job listing vs CV fit audit with market and community context

[rosspeili]

Skill Name

office/application_auditor

What should this skill do?

Add a skill that helps agents produce a structured hiring fit audit: how a CV matches public job listing(s), what is missing, optional market comparison, optional community tips, and actionable amendments (CV + cover letter points).

Proposed identity

Field	Value
Category	office (preferred) or auditing — confirm with maintainers before opening PR
Skill ID	office/application_auditor (or auditing/application_auditor if category approved)
Class	ApplicationAuditorSkill

User inputs (conceptual)

• Public URL(s) of job listings (HTML the user can access).
• CV as public URL, local file path, or pasted text (PDF/DOCX in v1 if feasible).
• Optional: run market research, run Reddit insights, region/seniority hints.

Skill outputs (conceptual)

• Job profile — role, seniority, must-have/nice-to-have, stack, location, compensation hints, evidence quotes.
• CV profile — experience, skills, achievements, evidence quotes.
• Fit analysis — fit band/score, matches, gaps, what the CV lacks to strengthen the application.
• Market context (v2, optional) — similar roles, salary/difficulty signals via a search API (no login-wall scraping).
• Community tips (v2, optional) — themes from Reddit API threads (sourced IDs; tips only).
• Recommendations — prioritized CV amendments and cover letter bullet points (structured JSON; host agent writes final prose).

Recommended architecture (Skillware)

Use a staged pipeline in one skill with a mode parameter, plus instructions.md playbook for the host agent:

Mode (v1)	Role
parse_job	Fetch URL → structured job_profile
parse_cv	File/path/text → structured cv_profile
analyze_fit	Compare profiles → gaps, fit, amendments, cover letter points

Mode (v2)	Role
market_signals	Web search API → comparable roles / comp context
community_insights	Reddit API → hiring tips themes
full_audit	Run full pipeline → single audit_report object

• Python: ingest only — httpx, HTML extraction (trafilatura / beautifulsoup4), CV PDF via pymupdf, optional python-docx; SSRF limits, size caps, no credential scraping.
• LLM inside skill: fixed prompts, JSON-schema outputs for extract/compare only (like synthetic_generator / MiCA evaluator pattern); model_provider + env keys.
• Host agent (any model): formats the user-facing report and polishes cover letter from skill JSON — skill does not own final presentation.

Do not build LangGraph/CrewAI designs; keep a portable Skillware bundle per CONTRIBUTING and templates/python_skill/.

Safety

• CV = PII: no persistent storage, no logging bodies, temp files only. (You can use Skillware for that)
• Public URLs user provides; respect ToS; no LinkedIn/login scraping in v1. (You can use Skillware for that)
• Reddit: official API only (praw + OAuth), rate limited.
• Disclaimer: not career, legal, or hiring advice; no outcome guarantees.

Phasing

• v1: parse_job, parse_cv, analyze_fit.
• v2: market + Reddit + optional full_audit.

Environment

Variable	When
GOOGLE_API_KEY / ANTHROPIC_API_KEY	LLM extract/compare
TAVILY_API_KEY or BRAVE_SEARCH_API_KEY	v2 market (implement one)
REDDIT_CLIENT_ID, REDDIT_CLIENT_SECRET	v2 Reddit

Could be using Google Search Engine API or simple requests/beautifulsoup with extra layers?

Fetch/extract-only paths must run in tests without API keys (mocked HTTP/LLM).

Done when

• Bundle under skills/office/application_auditor/ (or skills/auditing/... if category approved).
• docs/skills/application_auditor.md + Usage Examples (five providers).
• Row in docs/skills/README.md.
• test_skill.py, issuer validation, SkillLoader.load_skill(...) works.

Ideal Inputs & Outputs

Inputs (execute)

Parameter	Type	Required	Purpose
mode	string	Yes	parse_job | parse_cv | analyze_fit (v2: market_signals, community_insights, full_audit)
job_url	string	For parse_job	Public job listing URL
cv_source	string	For parse_cv	url or file_path
cv_text	string	Optional	Pasted CV text instead of file/URL
job_profile	object	For analyze_fit	From parse_job
cv_profile	object	For analyze_fit	From parse_cv
model_provider	string	Optional	gemini | anthropic | ollama for internal LLM steps
run_market_research	boolean	Optional	v2
run_reddit_insights	boolean	Optional	v2
region	string	Optional	e.g. EU, US for market queries

Outputs (by mode)

parse_job → job_profile

• title, company, seniority, must_have[], nice_to_have[], responsibilities[], location, compensation_signals[], source_url, evidence_spans[] (field, quote).

parse_cv → cv_profile

• skills[], roles[] (title, dates, highlights), education[], evidence_spans[].

analyze_fit →

• fit_band (strong | moderate | weak), fit_score (0–1),
• matched_requirements[], gaps[] (requirement, severity, suggestion),
• cv_amendments[] (section, action, content_hint),
• cover_letter_points[],
• evidence[] (claim, job_quote, cv_quote).

v2: market_summary, reddit_tips (with thread_ids), merged audit_report for the host agent to render.

Targeted Models (if applicable)

Model Agnostic (All)