Agent Skill Radar

Agent Skill Radar

Use this skill to find project opportunities from GitHub evidence instead of brainstorming from scratch.

Workflow

1. Run the radar CLI.
2. Read the top-scoring repos and issue samples.
3. Pick one opportunity with a narrow companion-tool shape.
4. Generate a build brief, fake-door probe, README angle, and launch post.
5. Save the report under reports/ and keep the data under data/.

Commands

Run the default daily scan:

agent-skill-radar run

Collect a custom scan:

agent-skill-radar collect \
  --query "agent memory coding assistant stars:>100 created:>=2025-01-01" \
  --limit-per-query 12 \
  --issue-limit 6 \
  --out data/custom.json

Render a report:

agent-skill-radar report --input data/custom.json --out reports/custom.md

Interpretation Rules

• Treat build-now as a proof-of-concept signal, not permission to build a large product.
• Prefer a companion tool, plugin, skill pack, template, benchmark, or CLI wrapper over a fork.
• Avoid topics where the only signal is stars. Look for issue comments, reactions, recurring workarounds, and ecosystem gaps.
• For giant repositories, build around the ecosystem edge: docs, config, migration, compatibility, monitoring, security, or publishing workflows.
• If the same gap appears across 3 or more repos, promote it to a standalone project idea.

Output Template

For each shortlisted idea, write:

Project:
Target user:
Observed demand:
Repo evidence:
Why upstream will not solve it:
Smallest useful artifact:
Distribution wedge:
Two-week scope:
Fake-door question:

Probe Template

Use a neutral validation comment. Do not pitch too early.

I have seen this come up in a few agent-tooling repos. My current workaround is <specific workaround>.

I am considering a small standalone <CLI / skill / MCP server / template> that does <specific job> without requiring upstream changes.

Before building it: would this solve your case, or is the real blocker different?