Turn “push it to GitHub” into a release people can find, understand and try.
An agent skill for GitHub launch preparation, search-aware repository metadata, evidence-backed READMEs, and a reviewed upload handoff. It works on an existing project; it is not an app builder or a promise of viral growth.
中文说明 · Read the skill · Research and alternatives · Local evidence
| Input | Output |
|---|---|
| A working local project | A concrete task/value statement, not a new invented product |
| Real search questions | A dated query-to-feature map with unknown demand kept unknown |
| Current project evidence | Factual README/About/topics and limitations |
| Approved public snapshot | Local hygiene report, file hashes and upload instructions |
| Real post-launch observations | A reviewed, scoped lesson proposal—not made-up growth wisdom |
Use github-discoverability on this repository.
Prepare its GitHub launch, but do not publish or change application code.
Research three close alternatives, improve the README and proposed metadata,
check the first-use instructions, and produce an exact upload handoff.
Mark every untested claim. Do not promise stars.
Upload only is a separate mode. It checks and publishes an already authored, explicitly approved file set; it does not tell the agent to build a new product.
Inspect SKILL.md and its resources first. The package follows the Agent Skills directory format. Real client activation and live distribution tests are listed separately in validation.
After this repository is publicly available, the intended third-party CLI route is:
npx skills add frabcd/github-discoverability-skill --skill github-discoverabilitySelect your agent in the installer. This route uses the external skills CLI and network access; its documented behavior is not an end-to-end test of this release. A release agent must test and record the actual remote route before claiming it works.
For local use, copy the entire skills/github-discoverability folder to a
project's .agents/skills/ for Codex or .claude/skills/ for Claude Code, avoiding
an existing skill with the same name. These are documentation-based paths;
client behavior can change and has not been exercised here.
Python 3.10+; no pip dependencies, credentials or network needed.
python skills/github-discoverability/scripts/preflight.py . --json
python scripts/demo.py
python -m unittest discover -s tests -vThe demo checks synthetic before/after fixtures. It is not a customer case or evidence of improved search rank. The checker does not write files or execute the target project. It does not scan Git history, every secret type, all media, or external links, and it does not judge whether a README is persuasive.
Before the repository identity is bound, the package audit correctly reports REVIEW. That is not a hidden failure: see publishing instructions.
GitHub's documented default repository search matches name, description and
topics; README search is available with in:readme. The skill checks these as
separate surfaces. It does not know GitHub's ranking formula, keyword volume, or
the share of users coming from search. See sources.
Do not use unrelated topics, fake metrics, auto-stars, promotional PR spam, or hidden keywords. A successful upload is not evidence of adoption.
Generic GitHub publishing and launch skills already exist, including close alternatives with privacy review and adoption feedback. This is an original, compact implementation—not a claimed new category. Read the dated comparison before choosing it.
The local helper needs only Python. The surrounding agent can have model costs and provider data processing; this package makes no fully-offline agent claim. The optional publication path needs authenticated GitHub access and explicit permission. A clean audit is never permission to publish.
Report a reproducible installation failure or a consented, sanitized release observation. See contribution rules. “It got stars, therefore this caused them” is not adequate evidence. No private user analytics are bundled.
Version 0.1.0. Locally authored and tested as recorded; no external user study, real launch outcome, search-rank result or 5K-star achievement is claimed.
MIT. Research sources are linked, not copied into the skill. Not affiliated with GitHub, OpenAI, Anthropic or Vercel.