VecGrep Action

A GitHub Action for semantic code search, PR validation, and architectural enforcement using VecGrep.

Modes

Mode	Description
index	Index the codebase and report stats
search	Run a semantic query and print results
validate	Search for a pattern — fail if matches found (fail_on_match) or not found (fail_on_no_match)
comment	Post search results as a PR comment
duplicate	Detect semantically similar code chunks across the codebase

Inputs

Input	Required	Default	Description
mode	No	search	Operation mode (see table above)
query	Depends	—	Semantic search query (required for search, validate, comment, duplicate)
path	No	.	Path to the codebase to index/search
top_k	No	8	Number of results to return
min_score	No	0.7	Minimum similarity score threshold (0.0–1.0)
fail_on_match	No	false	Exit 1 if any results are found above min_score
fail_on_no_match	No	false	Exit 1 if no results are found above min_score
comment_header	No	VecGrep Semantic Search Results	PR comment heading (comment mode)
github_token	No	—	GitHub token for posting PR comments

Outputs

Output	Description
results	Search results as a JSON array
match_count	Number of results above the score threshold
index_stats	Indexing summary (index mode)

Examples

Index on every PR

- uses: VecGrep/action@main
  with:
    mode: index
    path: .

Enforce architectural rules

Fail if raw SQL queries are found; require authentication logic to exist:

- uses: VecGrep/action@main
  with:
    mode: validate
    query: "raw SQL query string concatenation"
    min_score: "0.80"
    fail_on_match: "true"

- uses: VecGrep/action@main
  with:
    mode: validate
    query: "user authentication token verification"
    min_score: "0.75"
    fail_on_no_match: "true"

Comment on PRs with related code

- uses: VecGrep/action@main
  with:
    mode: comment
    query: "user authentication login session"
    top_k: "5"
    comment_header: "Related authentication code"
    github_token: ${{ secrets.GITHUB_TOKEN }}

Detect duplicate logic

- uses: VecGrep/action@main
  with:
    mode: duplicate
    min_score: "0.92"
    fail_on_match: "true"

Full example workflows are in the examples/ directory.

Docker

The action runs in a Docker container and can also be used standalone:

docker build -t vecgrep-action .

# Search
docker run --rm \
  -v $(pwd):/workspace \
  -e INPUT_MODE=search \
  -e INPUT_QUERY="database connection setup" \
  -e INPUT_PATH=/workspace \
  vecgrep-action

# Validate — exit 1 if pattern found
docker run --rm \
  -v $(pwd):/workspace \
  -e INPUT_MODE=validate \
  -e INPUT_QUERY="hardcoded password secret api key" \
  -e INPUT_PATH=/workspace \
  -e INPUT_MIN_SCORE=0.82 \
  -e INPUT_FAIL_ON_MATCH=true \
  vecgrep-action

License

MIT