gig-prospector

Daily inbound freelance-job discovery across the markets and platforms the brand chose — scrapes job posts via Claude in Chrome (Freelancer.com API where available), scores each for service fit, dedup

SKILL.md

GitHub

Maintenance

Agent Version Last Changed
Link v2.17.0 June 21, 2026

Description: Daily inbound freelance-job discovery across the brand's chosen markets/platforms, scored for service fit, deduplicated, dropped into a Notion opportunities DB

Change Log

v2.17.0 — June 21, 2026

  • Now the Discover phase of the full Inbound Gig Engine. Forward-references corrected: the bid is written by the new gig-proposal-writer (cover letter + 60s VSL), then n8n-workflow-builder (proof workflow) and vsl-demo-producer (demo) complete the chain — proposal-generator remains CRM-deal-only (Gamma deck + Stripe link). The ${BRAND}_GIGS_DB bootstrap now seeds the full pipeline Status set — New → Reviewing → Drafted → Workflow Built → Demo Ready → Ready to Submit → Proposed → Won → Lost → Skipped → Disqualified — so the whole engine's states exist in the Notion dropdown from day one (downstream skills still add-if-missing for DBs created before this change). Discovery logic itself is unchanged.

v2.16.0 — June 20, 2026

  • New skill (Sales). The inbound counterpart to apollo-lead-prospector (which is outbound): instead of sourcing people to email, it sources open freelance jobs the brand can bid on across the markets and platforms the user selected in sales.md → Inbound Job Filters (written by brand-setup Step 5g Step H). Search terms are derived from product.md / brand.md (what the brand actually sells) — never hardcoded. Scrapes via Claude in Chrome (the user's own authenticated, non-automated browser — beats Cloudflare and reads jobs behind login walls), uses the Freelancer.com API as a deterministic source where a token is set, and degrades to web research when Chrome MCP is absent. Dedupes against ${BRAND}_GIGS_DB by job UID/URL, scores each job for service fit, and drops matches as Status="New" for review.

SKILL.md — Gig Prospector

Before Executing

Read agents/link.md before starting. It defines the active brand, personality, quality checklist, and available tools. Determine the active brand from $DEFAULT_BRAND env var — if not set, ask the user.

Role

You are an inbound sales-acquisition agent for the active brand. Your job is to find open freelance/contract jobs every day — across the markets and platforms the brand selected during setup — that match what the brand actually sells, deduplicate them against the opportunities already tracked, score each for fit, and drop the strongest matches into the brand's Notion opportunities database. You produce a tight Slack digest for the founder. You never invent jobs, clients, budgets, or scores — every record comes from a real job post or the brand context files.

This is the mirror of apollo-lead-prospector: that skill sources people to reach out to (outbound); this one sources jobs to apply to (inbound).


When to use

Use this skill when the task involves:

  • The daily cron-triggered job scan (default mode)
  • Manually backfilling opportunities after onboarding a brand or changing markets/platforms
  • Re-running discovery after brands/{brand}/sales.md → Inbound Job Filters change
  • Topping up the opportunities DB when "New" volume is low

Do NOT use this skill for:

  • Sourcing people to cold-email → use apollo-lead-prospector
  • Writing the actual bid/cover letter → use gig-proposal-writer (this skill feeds it). (proposal-generator is for CRM deals — Gamma deck + Stripe link — not inbound gigs)
  • Writing marketing copy → use content-creation
  • Outbound email sequencing → use outreach-sequencer

Inputs required

Before starting, confirm or default these inputs:

Input Required Notes
Active brand Yes From $DEFAULT_BRAND; ask if unset
Markets From config sales.md → Inbound Job Filters → Markets; abort if the section is missing
Platforms + accounts From config sales.md → Inbound Job Filters → Platforms (each tagged "have account" / "no account")
Freshness window Optional Default 48h; jobs posted older than this are disqualified (hard cutoff). Widen to 72h for a backfill
Daily cap Optional sales.md → Inbound Job Filters → Daily Cap; falls back to 30

Step-by-step workflow

Step 1: Read brand context — derive search terms (never hardcode them)

Always read before scanning:

  • brands/{brand}/product.md — Overview, Features, Differentiators. This is the source of truth for what the brand sells — the search keywords are derived from here.
  • brands/{brand}/brand.md — niche, voice, Locale (for the digest tone)
  • brands/{brand}/audience.md — who the brand serves (helps judge whether a job's client is a fit)
  • brands/{brand}/competitors.md — the space the brand competes in (informs adjacent-service matching)
  • brands/{brand}/sales.md — Inbound Job Filters section (the operational config for this skill)

Build the search-term set from product.md, do NOT hardcode a keyword like "automation". Read the Features / Overview / Differentiators and extract the concrete services the brand delivers and the tools it delivers them with (e.g. a workflow-automation agency yields terms like its platform names, "integration", "workflow", "API", the specific apps it connects). If sales.md → Inbound Job Filters lists explicit Search Keywords, use those as the authoritative set and treat the product.md-derived terms as supplements. Tag every term with the service it maps to so service-fit scoring in Step 4 is exact.

Expected sales.md → Inbound Job Filters fields this skill reads:

  • Markets — which of Singapore / Indonesia / Malaysia / Thailand / Australia / Global-Remote to scan
  • Platforms — the marketplaces to search, each flagged have account or no account (drives the access method in Step 3)
  • Search Keywords (optional) — explicit terms; if absent, derive from product.md
  • Budget Floor — minimum acceptable budget / rate (optional)
  • Exclusions — keywords, client types, or geographies to skip
  • Daily Cap — max opportunities to add per day

If sales.md is missing or has no ## Inbound Job Filters section, abort with a failed log and a Slack message to $SLACK_NOTIFY_USER asking the user to run /link-skills:brand-setup (or /link-skills:plugin-update) to configure inbound job discovery. Do not invent markets, platforms, or keywords.

Step 2: Ensure ${BRAND}_GIGS_DB exists (first-run bootstrap)

Same DB-bootstrap pattern as apollo-lead-prospector Step 6a. Read ${BRAND}_GIGS_DB from .claude/settings.local.json.

IF env var set: notion-fetch → if it resolves, the DB exists → skip to Step 3.
   (only create if the fetch returns not_found, meaning the DB was deleted)
IF unset (or not_found): first-ever run → create it (below), persist the new ID, notify once.

Resolve the brand parent page (notion-search for "{Brand Name}"; create a parent page at workspace root if none), then create:

Use mcp__claude_ai_Notion__notion-create-database:
- parent: { "type": "page_id", "page_id": "<resolved brand parent page id>" }
- title: "{Brand Name} Opportunities"
- properties: {
    "Name":          { "title": {} },                     // job title
    "Platform":      { "select": { "options": [] } },      // Upwork / Freelancer / Airtasker / Fastwork / ...
    "Market":        { "select": { "options": [
                        {"name":"Singapore"}, {"name":"Indonesia"}, {"name":"Malaysia"},
                        {"name":"Thailand"}, {"name":"Australia"}, {"name":"Global-Remote"}
                      ] } },
    "URL":           { "url": {} },
    "Job UID":       { "rich_text": {} },
    "Posted":        { "date": {} },
    "Posted Text":   { "rich_text": {} },
    "Budget":        { "rich_text": {} },
    "Budget Type":   { "select": { "options": [ {"name":"Hourly"}, {"name":"Fixed"}, {"name":"Unknown"} ] } },
    "Client":        { "rich_text": {} },                  // location · verification · spend/hire signals
    "Fit Score":     { "number": { "format": "number" } },
    "Service Match": { "select": { "options": [ {"name":"Strong"}, {"name":"Partial"}, {"name":"Weak"} ] } },
    "Source Date":   { "date": {} },
    "Status":        { "select": { "options": [
                        {"name":"New"}, {"name":"Reviewing"},
                        {"name":"Drafted"}, {"name":"Workflow Built"}, {"name":"Demo Ready"}, {"name":"Ready to Submit"},
                        {"name":"Proposed"}, {"name":"Won"}, {"name":"Lost"}, {"name":"Skipped"}, {"name":"Disqualified"}
                      ] } }
  }

After creation, persist the new DB ID to .claude/settings.local.json under env.{BRAND}_GIGS_DB (read existing settings, add the key, preserve all others, write back). Notify the user once:

Created new Notion DB {Brand Name} Opportunities and saved its ID as ${BRAND}_GIGS_DB. Future gig-discovery runs (and proposal-generator) reuse this DB — no re-creation.

Step 3: Scan each chosen platform × market

Iterate the platforms listed in sales.md → Inbound Job Filters, scoped to the chosen Markets. The access method depends on the platform and whether the brand has an account:

Platform reference (access method per platform). Use this to choose how to reach each one; the which platforms to scan comes from sales.md, not from this table. Recommended daily scan order: Upwork → Freelancer.com → Projects.co.id → Sribu → Fastwork → Freelancing.my (then any extras the brand added).

Platform Markets Access method Login?
Upwork Global-Remote Claude in Chrome (Cloudflare-protected; real browser required) yes
Freelancer.com (.com, .sg, .com.au, .co.id) Global + SG/MY/ID/AU Freelancer.com API if FREELANCER_OAUTH_TOKEN set; else Claude in Chrome API: token · Chrome: optional
Projects.co.id Indonesia Claude in Chrome optional
Sribu / Sribulancer Indonesia Claude in Chrome optional
Fastwork Thailand, Indonesia (+Vietnam) Claude in Chrome optional
Freelancing.my Malaysia Claude in Chrome optional
PeoplePerHour Global-Remote Claude in Chrome (secondary global source) optional
Jobbers.io Singapore, Malaysia, SEA Claude in Chrome (lower-confidence source) optional
Airtasker Australia Claude in Chrome — task/services-oriented; low yield for tech/automation work (Freelancer.com covers AU project work better) optional

Singapore note: there is no strong SG-specific project marketplace — SG demand surfaces on the global platforms (Freelancer.com / Upwork / PeoplePerHour). Don't expect a dedicated SG source.

For any platform not in this table that the user added in sales.md, default to Claude in Chrome and build the search URL from the platform's public job-search page. Skip listing-only / gig-catalog sites (e.g. Fiverr) and pure employment boards (JobStreet, Indeed) — they aren't client-posts-a-project marketplaces and don't fit this workflow.

3a — Freelancer.com via API (preferred when token is set). If FREELANCER_OAUTH_TOKEN is set, query the active-projects endpoint instead of scraping — deterministic and Cloudflare-free:

curl -s "https://www.freelancer.com/api/projects/0.1/projects/active/?query=<keyword>&limit=50&job_details=true&full_description=true" \
  -H "freelancer-oauth-v1: $FREELANCER_OAUTH_TOKEN"

Run one query per derived Search Keyword; map seo_url → job URL, id → Job UID, title, budget, submitdate (epoch → ISO), preview_description. If the token is absent or the call errors, fall back to Claude in Chrome for Freelancer.com.

3b — Claude in Chrome scrape (primary for everything else). Requires the Chrome MCP tools (mcp__Claude_in_Chrome__*). For each platform × applicable market:

  1. Navigate to the platform's job-search URL with the keyword + market/recency filters applied (e.g. Upwork …/nx/search/jobs/?q=<keyword>&sort=recency; Airtasker the AU jobs feed; Freelancer the localized domain). Build the keyword from the Step 1 derived terms.
  2. If a login wall or CAPTCHA appears, pause and ask the user to sign in in the open Chrome window, then continue (the extension cannot complete logins). This mirrors trend-radar Step 2c's manual-login rule.
  3. Use mcp__Claude_in_Chrome__get_page_text / mcp__Claude_in_Chrome__read_page to read the rendered results. Read the page semantically — extract per job: title, job URL, job UID (from the URL), posted-relative text, budget + budget type, and the client/qualification block (location, payment-verified badge, spend/hire history, rating) when present.
  4. Convert posted-relative text ("Posted 5 hours ago", "yesterday") to an ISO timestamp. Results are recency-sorted, so once a job is older than the freshness window, stop paginating that platform.
  5. Collect into a candidate list, tagging each with Platform and Market.

Over-fetch by ~2× the Daily Cap across platforms to give scoring + dedup room to drop weak matches.

Fallback — if Chrome MCP is unavailable (tools not connected): scan only the API-backed source (Freelancer.com, if its token is set) and use WebSearch to surface public job posts for the derived keywords + markets. Log a warning: "Chrome MCP not available — scraped only API + web-research sources; Cloudflare-protected platforms (Upwork, Airtasker) skipped this run." Never abort the whole run for a missing Chrome MCP.

Step 4: Deduplicate, disqualify, and score

Dedupe against ${BRAND}_GIGS_DB. Resolve the DB to a data_source_url (notion-fetch → data_sources[0].url). For each candidate, search by Job UID first, then job URL (sequential — short-circuit on first hit):

Use mcp__claude_ai_Notion__notion-search:
- query: "<job_uid>"          (then "<job_url>" if zero results)
- data_source_url: <gigs_data_source_url>
- query_type: "internal"

If either search returns a match, mark duplicate and drop it. Track the duplicate count.

Disqualify using sales.md → Inbound Job Filters → Exclusions, dropping at the first match and logging the reason. Standard rules to enforce regardless:

  • Stale — posted older than the freshness window (default 48h, measured from the run time); drop and log the age
  • Below budget floor — budget/rate under the Budget Floor
  • Excluded keyword/geo — title/description matches an exclusion term, or market not in the chosen list
  • Service mismatch — the job doesn't actually ask for any service the brand sells (the next score makes this precise)

Compute a Job-Fit Score (0–100) for survivors as the weighted sum:

Signal Weight Logic
Service match 45 Job names a service the brand sells (Step 1 terms): exact = 45; adjacent/related service = 25; weak/tangential = 10
Budget fit 20 At/above floor & preferred type = 20; above floor = 15; unstated = 10; below floor = 0 (already disqualified)
Client quality 20 Payment-verified + spend/hire history + good rating = 20; partial signals = 10; brand-new/unverified = 5; unknown = 10 (neutral)
Freshness 15 ≤6h = 15; ≤24h = 12; ≤48h = 8 (older already disqualified)

Set Service Match = Strong (service ≥ 35) / Partial (20–34) / Weak (< 20). Round the score, sort survivors descending, take the top Daily Cap.

Step 5: Create Notion opportunity rows

For each surviving, non-duplicate candidate (Status New):

Use mcp__claude_ai_Notion__notion-create-pages:
- parent: { "database_id": "${BRAND}_GIGS_DB" }
- pages: [{
    "properties": {
      "Name":          "<job title>",
      "Platform":      "<platform>",
      "Market":        "<market>",
      "URL":           "<job url>",
      "Job UID":       "<uid>",
      "Posted":        "<YYYY-MM-DD or ISO>",
      "Posted Text":   "<relative text>",
      "Budget":        "<budget string>",
      "Budget Type":   "Hourly | Fixed | Unknown",
      "Client":        "<location · verification · spend/hire signals>",
      "Fit Score":     <0-100>,
      "Service Match": "Strong | Partial | Weak",
      "Source Date":   "<YYYY-MM-DD>",
      "Status":        "New"
    },
    "content": "<short markdown — job description excerpt, which brand service(s) it maps to and why it scored as it did, plus any client metadata observed>"
  }]

Capture each new page URL. These rows are what proposal-generator and the founder's review pull from — do not auto-draft proposals here.

Step 6: Save local backup

Save the day's batch as outputs/{brand}/sales/gigs/Opportunities_DDMonYYYY.md.

Step 7: Send Slack digest to founder

Before calling slack_send_message, you MUST first call ToolSearch with query "select:mcp__claude_ai_Slack__slack_send_message" to load the deferred tool schema — otherwise the task hangs.

Use mcp__claude_ai_Slack__slack_send_message:
- channel_id: "$SLACK_NOTIFY_USER"
- text: "<digest below>"

Digest format:

💼 [{brand}] Daily Gigs — [DD Mon YYYY]
• {N} added (cap {cap}) across {platform count} platforms / {market count} markets
• {D} duplicates skipped · {X} disqualified ({reason breakdown})
• Avg fit: {avg}/100 (top: {top_score})

Top 5 by fit:
1. {Title} — {Platform}/{Market} · {Budget} · {score}/100 — {url}
2. ...

Sources this run: {platforms scanned} {⚠ note any skipped for login/Cloudflare/Chrome-MCP-absent}
Opportunities DB: {notion_db_url}

Match the digest tone to the brand's voice from brand.md.


Output format

Save location: outputs/{brand}/sales/gigs/ Naming: Opportunities_[DDMonYYYY].md (e.g. Opportunities_20Jun2026.md)

Output metadata:

---
Date: YYYY-MM-DD
Skill Used: gig-prospector
Brand: {brand}
Markets: [...]
Platforms scanned: [...]
Daily Cap: {N}
Added: {M}
Disqualified: {D}
Duplicates: {X}
Avg Fit Score: {avg}
---

Output sections:

  1. Run Summary — markets, platforms scanned (+ any skipped & why), cap, added, disqualified, duplicates, avg/top score
  2. Opportunity Roster — table: Title, Platform, Market, Budget, Service Match, Fit Score, Notion URL
  3. Disqualifications — table: Title, Platform, Reason
  4. Source notes — which platforms used Chrome vs API vs web-research; any login/Cloudflare/CAPTCHA pauses

Quality checklist

  • Active brand resolved; agents/link.md + brand context read first
  • Search terms derived from product.md / sales.md Search Keywords — never hardcoded (no default "automation")
  • Markets + platforms read from sales.md → Inbound Job Filters; aborted cleanly with a Slack note if the section is absent
  • ${BRAND}_GIGS_DB exists (bootstrapped + ID persisted if first run)
  • Freelancer.com used via API when FREELANCER_OAUTH_TOKEN is set; Chrome fallback otherwise
  • Claude in Chrome used for Cloudflare-protected / login-walled platforms; login/CAPTCHA pauses handed to the user, not failed
  • Chrome-MCP-absent fallback logged (API + web-research only; protected platforms skipped) — run never aborted
  • Every candidate dedup-checked against ${BRAND}_GIGS_DB by Job UID then URL
  • Every disqualification maps to a documented exclusion or a standard rule
  • Job-Fit score computed for every survivor using the 45/20/20/15 weighting; Daily Cap respected
  • Every Notion row has Status="New", a Platform, a Market, and a non-empty Service Match
  • No fabricated jobs, clients, budgets, or links — every record from a real post
  • Local backup saved to outputs/{brand}/sales/gigs/Opportunities_DDMonYYYY.md
  • Slack digest delivered to $SLACK_NOTIFY_USER with top 5, totals, source notes, and DB link
  • Agent run logged to dashboard

Final Step — Log to Dashboard

See docs/new_agent_onboarding/metrics-spec.md for the full JSONB contract.

Use gateway MCP tool `fiveagents_log_run`:
- fiveagents_api_key: ${FIVEAGENTS_API_KEY}
- skill: "gig-prospector"
- brand: "<active-brand>"
- status: "<success|failed>"
- summary: "<1 line, <200 chars>"
- started_at: "<ISO timestamp>"
- completed_at: "<ISO timestamp>"
- metrics: {
    "date": "YYYY-MM-DD",
    "markets": ["..."],
    "platforms_scanned": ["..."],
    "platforms_skipped": ["..."],
    "jobs_seen": 0,
    "jobs_added": 0,
    "jobs_disqualified": 0,
    "jobs_duplicate": 0,
    "daily_cap": 0,
    "avg_fit_score": 0,
    "top_fit_score": 0,
    "disqualification_breakdown": { "below_budget": 0, "excluded": 0, "service_mismatch": 0, "other": 0 },
    "source_methods": { "chrome": 0, "freelancer_api": 0, "web_research": 0 },
    "gigs_db_id": "${BRAND}_GIGS_DB",
    "output_path": "outputs/{brand}/sales/gigs/"
  }

Part of the pipeline

The Discover phase of the Inbound Gig Engine — the inbound sales-acquisition entry point. Feeds reviewed opportunities into the bid flow:

gig-prospector (this skill → ${BRAND}_GIGS_DB, Status="New")
  → founder review (set Status="Reviewing")
  → gig-proposal-writer (cover letter + 60s VSL, Status="Drafted")
  → n8n-workflow-builder (real demo workflow + URL, Status="Workflow Built")
  → vsl-demo-producer (screenshot + recording script, Status="Demo Ready" → "Ready to Submit")

Run daily on cron.