Interview the user about what they're looking for, research a ranked company list, and write it into their Target Companies tracker — Notion or local files, whichever they use — safe to re-run to refresh the list without creating duplicates.
Instructions
Determine where the user's tracker lives:
Search broadly for an existing local tracker first — e.g. find . -maxdepth 4 -iname "target-companies.csv" — rather than only checking the default ./job-search-tracker/ folder, since the user may have chosen a different location when it was set up. If found, use local files at that path.
Else if a Notion database titled "Target Companies" exists, use Notion.
Else tell the user: "You don't have a target company tracker yet — let me set one up first," then invoke the notion-connector skill (which will itself ask whether to use Notion or local files). Continue once it's created, using whichever storage the user chose.
Ask the following, one at a time (wait for each response):
"What area or domain are you targeting?" (industry or function, e.g. "fintech", "AI infrastructure", "gaming")
"Any location preference — a specific city, hybrid, or fully remote?"
"What stage of company: startup (seed/Series A), growth (Series B+), public/enterprise, or any?"
"What role or level are you targeting?" (e.g. "Senior PM", "Staff Engineer")
"How many companies should I find? (default 20 if no preference)"
"Any companies I should exclude — already applied, not interested?" (optional, can skip)
Run web searches to build a candidate list matching the stated criteria. For each candidate company capture: name, category/domain, stage, location, careers page URL if found, target level (the user's stated answer, applied to all candidates since it's a search criterion, not something that varies per company), and a one-line "why it fits" tied to the user's stated criteria. Exclude anything the user asked to exclude.
Group the candidates into three tiers based on fit against the stated criteria:
Dream — matches all criteria closely, most desirable
Strong Fit — matches most criteria well
Backup — plausible but a weaker match on one or more criteria
Present the full list as a markdown table (Company | Tier | Category | Stage | Location | Why It Fits) to the user before writing anything. Ask: "Want me to save this to your tracker, or should I adjust anything first?"
On approval, write the results — using whichever storage was determined in step 1:
For each candidate, look up an existing row by company name (case-insensitive match) — query Notion by property value, or read the local CSV and match on the Company column.
Match found: update Tier, Category, Stage, Location, Careers URL, Target Level, Why It Fits. For Notes, append the new note rather than overwrite existing content. Never touch Warm Contact or Status — those are user-maintained.
No match: add a new row (a new Notion page, or a new CSV line) with Status = "Not Started" and all captured fields filled in, including Target Level.
For local files, read the whole CSV with a real CSV parser, apply the match/update/append logic in memory, and write the full file back once with a real CSV writer (RFC 4180 quoting/escaping) — never append blindly without checking for an existing row first, and never build rows by joining fields with commas manually, since "Why It Fits" is free text that will often contain one.
Report a summary: "Added N, updated N." with the database link, or the local file path.
Rules
Never invent a company or fact about it — if web search can't confirm something (e.g. stage), leave that field blank rather than guessing.
If fewer solid matches are found than requested, say so explicitly and return what was found — never pad the list with weak fits to hit the count.
If a write fails partway through the batch, report exactly which companies succeeded and which didn't. Do not retry silently.
Never overwrite Warm Contact or Status on an existing row — those fields are user-maintained.