mirror of
https://github.com/Significant-Gravitas/AutoGPT.git
synced 2026-11-10 04:41:59 -10:00
Nicholas TindleandClaude Opus 5.5
f90049a17a
### Why / What / How
**Why:** Agents and AutoPilot can't answer "how many views and clones
did we get?". GitHub's official MCP server, which is in our catalog, has
no traffic tools; there's an open request for them
(github/github-mcp-server#3164). The nightly `github-repo-stats`
workflow was removed in #15169 (an internal collector now archives the
org's own repos), so nothing on the platform reads traffic for a user's
repository. It's on the growth and marketing analytics wish list, next
to PostHog (already in the MCP catalog), Hacker News, Google Analytics
and Search Console.
**What:** One read-only block in the GitHub family, Github Get
Repository Traffic. One run returns what a repository's Insights >
Traffic page shows.
**How:**
- Calls the four traffic endpoints at the same time (`asyncio.gather`)
through the existing `get_api`:
- `GET /repos/{owner}/{repo}/traffic/views?per=day|week`
- `GET /repos/{owner}/{repo}/traffic/clones?per=day|week`
- `GET /repos/{owner}/{repo}/traffic/popular/referrers` (top 10)
- `GET /repos/{owner}/{repo}/traffic/popular/paths` (top 10)
- **Inputs:** GitHub credentials (`repo` scope), `repo_url` (a
github.com link or `owner/repo`, checked by `normalize_repo_path`) and
`period` (`day` by default, or `week`).
- **Outputs:** `views`, `unique_visitors`, `views_by_period`, `clones`,
`unique_cloners`, `clones_by_period`, `referrers` and `popular_content`,
named after the Traffic page's labels. Each popular page also gets its
github.com `url`. Lists only, no per-item outputs: they are short and
read whole.
- **Errors:**
- A 403 saying the caller can't read traffic ("Must have push access to
repository", or "Resource not accessible by ..." for a fine-grained
token) becomes: "GitHub only shows traffic to people with push access to
owner/repo. Connect a GitHub account that can push to it; a fine-grained
token also needs the Administration (read) permission."
- A 404 says the repository wasn't found or the account can't see it.
- Other 403s (rate limits, SAML SSO, IP allow lists) and 401s keep
GitHub's message. A 401 reaches AutoPilot unchanged, so its reconnect
prompt still works.
- An invalid repository is a `BlockInputError`, raised before any
request.
- **Conventions:** each request is a static method the self-test mocks.
`run()` does the gather, error mapping and parsing, so the self-test
parses realistic payloads. `run()` isn't wrapped in a broad
`try/except`. Effect is `READ`, so AutoPilot runs it without asking.
Categories: Developer Tools and Marketing.
### Changes 🏗️
- `backend/blocks/github/traffic.py`: `GithubGetRepositoryTrafficBlock`
(`aa46e092-b04f-492a-ae13-bb920f1469c2`) and its `TrafficCount`,
`TrafficReferrer` and `PopularContent` output types.
- `backend/blocks/github/traffic_test.py` (29 tests) and
`traffic_effects_test.py` (pins `BlockEffect.READ`).
- Docs: `docs/integrations/block-integrations/github/traffic.md`, plus
the generated `README.md` and `SUMMARY.md` entries.
- No existing blocks, settings or env vars change.
- **What users need:** push (write) access to the repository. The GitHub
connection or a classic token needs the `repo` scope; a fine-grained
token needs the Administration repository permission (read). GitHub
keeps 14 days, so the docs suggest a daily run that appends to a Google
Sheet for longer history.
### Agents and large language models used
- Claude Code with Claude Opus 5.5
### Checklist 📋
#### For code changes:
- [x] I have clearly listed my changes in the PR description
- [x] I have made a test plan
- [x] I have tested my changes according to the test plan:
- [x] Block self-test passes via `execute_block_test`, and
`test_block.py -k GithubGetRepositoryTraffic --noconftest` gives 6
passed (self-test, effect checks, uuid4 id)
- [x] New and existing GitHub tests: 160 passed (`traffic_test.py`,
`traffic_effects_test.py`, `test_github_blocks.py`, `--noconftest`). The
new ones cover:
- URLs and the `per` parameter for each endpoint, and `get_api` built
with `convert_urls=False`
- all four requests in flight at once
- URL and `owner/repo` input, and invalid repositories and periods
rejected before any request
- parsing realistic, empty, missing and null fields, and page links
- error mapping: no push access, fine-grained token, 404, rate limit,
401
- [x] `test_block.py -k "declares_an_effect or ids_valid or effect_none
or irreversible" --noconftest`: 3037 passed, 83 skipped
- [x] `copilot/capabilities/registry_test.py` and
`integrations/provider_logos_test.py`: 146 passed. The copilot listing
is 297 characters, and copilot search ranks the block first for "how
many clones did our github repo get".
- [x] `ruff`, `isort` and `black` clean over the whole backend;
`pyright` 0 errors
- [x] `load_all_blocks()` loads all 624 blocks; `generate_block_docs.py
--check` reports all documentation in sync
- [x] Ran the block itself against GitHub with a real token that can
push to Significant-Gravitas/AutoGPT, weekly breakdown: 40,434 views
(19,507 unique visitors), 186,288 clones (8,347 unique cloners), two
weekly buckets each, 10 referrers and 10 pages with their links. Both
`owner/repo` and a github.com URL worked.
- [x] Errors from the same live run: `torvalds/linux` (no push access)
gave the push-access message, a repository that doesn't exist gave the
not-found message, and `not a repo` was rejected before any request.
- [ ] `test_block.py` against the Docker test stack (not run; the runs
above use `--noconftest`)
**Not verified:**
- The fine-grained token 403 wording ("Resource not accessible by
personal access token") is GitHub's usual message for a missing
permission. I didn't see it from this endpoint.
- In the live check the newest daily entry was yesterday (UTC). I don't
know whether GitHub ever includes the current, partial day.
#### For configuration changes:
- [x] `.env.default` is updated or already compatible with my changes
(no new settings)
- [x] `docker-compose.yml` is updated or already compatible with my
changes (no changes)
- [x] I have included a list of my configuration changes in the PR
description (under **Changes**). There are none.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- CURSOR_SUMMARY -->
---
> [!NOTE]
> **Low Risk**
> Additive read-only integration with isolated new code and tests; no
auth, persistence, or existing block behavior changes.
>
> **Overview**
> Adds a **read-only** GitHub block, **Github Get Repository Traffic**,
so workflows can pull the same metrics as **Insights → Traffic** (14-day
views/clones, referrers, popular paths).
>
> One run **fans out four REST calls in parallel** (`views`, `clones`,
`popular/referrers`, `popular/paths`) via the existing `get_api` helper
(`convert_urls=False`). Inputs are `repo_url` (normalized with
`normalize_repo_path`), optional `period` (`day`/`week` for
views/clones), and repo-scoped credentials. Outputs aggregate totals,
per-period breakdowns, top referrers, and popular pages with built
`github.com` URLs.
>
> **Error handling** maps traffic-specific 403s and 404s to clearer
`BlockExecutionError` messages; rate limits and 401s are left for
AutoPilot. The block declares **`BlockEffect.READ`** (Developer Tools +
Marketing categories).
>
> Also adds **unit/effect tests** and integration docs (`traffic.md`,
index entries in `README.md` / `SUMMARY.md`). No changes to existing
blocks or configuration.
>
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
f50bfd64e2. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->
Co-authored-by: Claude Opus 5.5 (Claude Code) <[email protected]>