Competitive Intelligence Agent

Competitive Intelligence Agent

Use this skill to operate a competitive-intelligence repo as an agent tool system, not just a report script. Prefer a reliable sample-first path, then opt into live CocoIndex only when credentials and Postgres are ready.

Workflow

1. Find the repo root. Look for mcp_server.py, local_intel.py, watchlist.json, and optional docker-compose.yml.
2. Inspect current state before editing: git status --short, rg -n "run_cocoindex_update|FastMCP|CocoIndex|COMPETITORS" ..
3. Use sample mode first unless the user explicitly asks for live ingestion.
4. For live mode, verify .env has COCOINDEX_DATABASE_URL, TAVILY_API_KEY, and OPENAI_API_KEY; never print secret values.
5. Keep generated briefs, dashboards, and JSON/CSV report artifacts out of commits unless the repo explicitly tracks samples.
6. Run focused tests before committing.

Sample Demo

Use this path for a reliable demo without API keys:

python3 local_intel.py --dashboard --slug demo
python3 agent_demo.py --slug demo-agent
python3 -m unittest test_local_intel.py

Expected outputs usually land in reports/:

• Markdown brief
• JSON/CSV event export
• static HTML dashboard
• deterministic agent transcript, if agent_demo.py exists

MCP Tool Demo

Start the MCP server from the repo environment:

python3 mcp_server.py

For client setup, prefer a repo-provided mcp-config.example.json. Use an absolute Python path and an absolute mcp_server.py path. If the repo is also packaged as a Codex plugin, keep plugin-specific local paths out of public docs.

Useful agent calls:

get_trending_competitors(limit=5)
search_events(query="enterprise customers", limit=5)
create_brief(slug="board-brief")
create_dashboard(slug="demo-dashboard")

Live CocoIndex

Use live ingestion only when explicitly requested. Start Postgres if the repo provides Docker support:

docker compose up -d postgres

Run with explicit per-call competitors instead of editing .env for every scenario:

run_cocoindex_update(
  live=true,
  competitors="Apple,Microsoft",
  max_results=2,
  event_query="(product launch OR partnership)",
  search_days_back=14
)

Then query the indexed data:

search_events(mode="cocoindex", competitor="Apple", limit=5)
get_trending_competitors(mode="cocoindex", days=7)
create_brief(mode="cocoindex", slug="apple-microsoft-live")
create_dashboard(mode="cocoindex", slug="apple-microsoft-live")

competitors may be a comma-separated string or a JSON array. Treat .env values as defaults, not fixed behavior.

Validation

For code changes, run:

python3 -m py_compile main.py mcp_server.py providers.py local_intel.py
python3 -m unittest test_local_intel.py

For live changes, add one small credential-gated smoke:

python3 live_demo_check.py \
  --slug live-smoke \
  --competitors "Apple" \
  --max-results 1 \
  --event-query "(product launch)"

Report whether live mode used sample data, local Postgres, or a remote database. Be explicit when live checks were skipped because credentials, Docker, network, or API quota were unavailable.

References

Read references/cocoindex-agent-demo.md when adding docs, README sections, plugin instructions, or MCP client examples for this workflow.