Skip to content

SearXNG engine roster in the template; source-plugin design note - #22

Merged
markramm merged 3 commits into
mainfrom
keep/searxng-and-plugin-design
Sep 29, 2026
Merged

markramm merged 3 commits into
mainfrom
keep/searxng-and-plugin-design

Conversation

@markramm

Copy link
Copy Markdown
Owner

Three commits that sat on local main, generic and safe to publish (Mark approved, 2026-09-29):

  • the SearXNG fallback engine roster in the tracked template (the secret key stays a placeholder; the real one is generated into the gitignored settings.local.yml), trimmed to the three engines that return results, with mojeek's network-level 403 documented;
  • docs/source-plugin-design.md, a design note for a source-plugin registry, which the extension-candidates analysis (docs: extension candidates ranked against live reporting (refs #7) #15) builds on.

Tests: 262 passed, 2 xfailed.

🤖 Generated with Claude Code

markramm and others added 3 commits September 29, 2026 11:00
… gitignored local override

The four default general engines (brave, google cse, duckduckgo, startpage) are durably blocked
for this instance. mwmbl/qwant/crowdview are what keep general search alive. Without this in the
template, a fresh clone inherits a roster where general web search is at zero.

(cherry picked from commit 68f2234d661df4f3cc56585c5896471b7b766bd8)
…t mojeek's network-level 403

(cherry picked from commit e7c88b93d4d7ec33bb8a14b2746cd29980e31247)
Proposal only, nothing built. Measures the current cost of adding a source
(five touchpoints, three of them edits to shared files; no registry exists;
main() carries nine hardcoded imports) and proposes a SourceSpec dataclass
plus a registry that cli.py iterates.

The finding that makes it cheap: the self-documentation Mark wants already
exists. corpus / not_searched / caveats / exact_match_supported are already
declared by five of seven sources -- as loose kwargs to run_source, so nothing
can enumerate them without running a search. Crossref already tells you its
title search is fuzzy across ~150M records and that a differently-worded title
would not be found. That is a better self-description than most MCP tool
definitions because it describes what the tool CANNOT tell you.

Includes scope discipline: no third-party entry-point loading in v1; searxng
will not fit a uniform runner (per-engine coverage); usaspending has four entry
points so one-verb-per-source is wrong; and the manifest must not become a place
to assert unverified coverage, since a wrong not_searched produces an
overclaimed absence.

Suggested first step is converting two sources -- one simple, one awkward --
rather than all nine.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
(cherry picked from commit b0969f37aa3da0d34715f9e9521073cdabfeac2e)
@markramm
markramm merged commit aef6849 into main Sep 29, 2026
3 checks passed
@markramm
markramm deleted the keep/searxng-and-plugin-design branch September 29, 2026 15:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant