Skip to content

Add scoped auto-import flow and automate daily run script - #5

Merged
simonperegrinvs merged 3 commits into
mainfrom
feat/auto-scope-discovery
Feb 18, 2026
Merged

simonperegrinvs merged 3 commits into
mainfrom
feat/auto-scope-discovery

Conversation

@simonperegrinvs

Copy link
Copy Markdown
Owner

Summary

  • add scoped filtering to agent discovery import and agent discovery ingest --auto-import
  • classify candidates as in-scope-core, in-scope-adjacent, or out-of-scope using include/exclude keywords + node-link score
  • reject out-of-scope candidates with explicit decisionReason and audit entries
  • expose new CLI controls:
    • --scope-keyword <keywords...>
    • --exclude-keyword <keywords...>
    • --min-scope-score <n>
    • --no-scope-filter
  • replace run-agent.sh prompt-driven flow with a deterministic CLI loop using discovery auto-import + scope filtering
  • update docs (CLI.md, README.md) and extend behavior tests for scope filtering

Why

Recent auto-import runs could pull noisy/off-topic papers when discovery lanes were broad. This adds a practical safety gate so automation remains useful without constant manual cleanup.

Validation

  • npm test (24 files / 117 tests) passes
  • npm run build passes
  • bash -n run-agent.sh passes
  • real flow test on temp graph (mixed query) imported in-scope psi candidates with scope classifications
  • real flow test on temp graph (noise query) showed outOfScope rejections (imported: 0, outOfScope: 5)

Notes

  • Semantic Scholar intermittently returned retry-limit failures during live testing; OpenAlex lane was used as fallback for deterministic verification.

@simonperegrinvs
simonperegrinvs merged commit e8ee98a into main Feb 18, 2026
1 check passed
@simonperegrinvs
simonperegrinvs deleted the feat/auto-scope-discovery branch February 18, 2026 14:09
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