ncalavera

ncalavera/llm-job-pipeline

Agent Claude Code

Nightly headless scoring subagent. Reads ONE payload file from vacancies/nightly/ /scorein/ and writes ONE result file to scoreout/. Spawned only by /jobs-night; never used interactively.

ncalavera/llm-job-pipeline

Command Claude Code

Add a source to the pipeline in one of three modes — company (auto-detect ATS, default), board (enable a built-in job board via JOBBOARDS), or vacancy/job (hand-add a single role that auto-scores in /jobs-new).

ncalavera/llm-job-pipeline

Command Claude Code

Send a daily Telegram digest of top unseen vacancies and poll button responses back into the database.

ncalavera/llm-job-pipeline

Command Claude Code

Scoring-quality check. Build a small golden set of your own vacancies labelled fits / doesn't-fit, score them with the CURRENT prompt, and get one agreement number plus a list of disagreements to fix. Any user can run it; the set is personal and never leaves your machine.

ncalavera/llm-job-pipeline

Command Claude Code

The one daily command. A Python driver runs the whole pipeline in a fixed order — validate → fetch → enrich → filter → score → verdicts → publish — with checkpoints, a live progress card, and a publish gate. You (the agent) only supply judgment at the gates it stops on: scoring, WANT-scoring companies, and like/pass…

ncalavera/llm-job-pipeline

Command Claude Code

Headless nightly scoring session. Invoked by scripts/nightlyrun.py as /jobs-night — one session per scoring gate. Reads payload files from /scorein/, fans them out to night-scorer subagents (file-in/file-out), saves after every wave of five, then stops. The wrapper resumes the driver; the session never runs it. Never…

ncalavera/llm-job-pipeline

Command Claude Code

View or edit your candidate profile — summary view, hard-filter rules, or broader section edits, all in one command. Source is config/userprofile.md.

ncalavera/llm-job-pipeline

Command Claude Code

Review hub for the vacancy pipeline. Bare = status menu; apply = deep liked-vacancy interview; archive = interactive low-score archival; vac = thin triage CLI. Auto-publishes after state changes (deploy only, never git).

ncalavera/llm-job-pipeline

Command Claude Code

Update the pipeline — pull latest code, refresh dependencies, apply pending DB migrations, summarise what changed.

ncalavera/llm-job-pipeline

Hook Claude Code

Runs before the agent uses a tool for Write, Edit, MultiEdit and NotebookEdit tool calls, executing night-write-fence.py via python3. From ncalavera/llm-job-pipeline.

ncalavera/llm-job-pipeline

Settings file Claude Code

Agent settings declaring 1 hook event (PreToolUse).

ncalavera/llm-job-pipeline

Instructions file CodexOpenCode

Instructions for ncalavera/llm-job-pipeline, covering agent instructions, read strategy.md first (hard rule), where the runbooks live, the scoring contract (any llm agent) and no direct-api key is a supported setup, not a defect.