semanticembed 0.7.3__tar.gz → 0.7.5__tar.gz

{semanticembed-0.7.3 → semanticembed-0.7.5}/.claude/skills/analyze/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: analyze
 description: Scan a repo for infrastructure files (docker-compose, k8s, terraform, GitHub Actions), extract edges, and run 6D structural risk analysis via SemanticEmbed.
-allowed-tools: Bash Read Glob
+allowed-tools: Read Glob
 ---
 
 # /analyze — Structural Risk Analysis

semanticembed-0.7.5/.github/CODEOWNERS

@@ -0,0 +1,5 @@
+# Changes to these paths need review from the owner once main has a ruleset
+# that requires code-owner review.
+*                   @jmurray10
+/.github/            @jmurray10
+/src/semanticembed/  @jmurray10

{semanticembed-0.7.3 → semanticembed-0.7.5}/.github/ISSUE_TEMPLATE/bug_report.yml

@@ -6,7 +6,7 @@ body:
     attributes:
       value: |
         Thanks for taking the time to report a bug. Please fill in the fields below.
-        For algorithm/encoding/scoring questions, email jeffmurr@seas.upenn.edu — that
+        For algorithm/encoding/scoring questions, email jeff.murray@alumni.upenn.edu — that
         runs server-side and isn't part of this repository.
 
   - type: input

{semanticembed-0.7.3 → semanticembed-0.7.5}/.github/ISSUE_TEMPLATE/config.yml

@@ -1,10 +1,10 @@
 blank_issues_enabled: false
 contact_links:
   - name: Algorithm / encoding / scoring questions
-    url: mailto:jeffmurr@seas.upenn.edu
+    url: mailto:jeff.murray@alumni.upenn.edu
     about: |
       The 6D encoding algorithm runs server-side and isn't part of this repository.
       Email directly for questions about encoding math, risk thresholds, or scoring rules.
   - name: License / commercial inquiries
-    url: mailto:jeffmurr@seas.upenn.edu
+    url: mailto:jeff.murray@alumni.upenn.edu
     about: Pricing, paid-tier license keys, and commercial use questions.

{semanticembed-0.7.3 → semanticembed-0.7.5}/.github/ISSUE_TEMPLATE/feature_request.yml

@@ -11,7 +11,7 @@ body:
         - **Client improvements** — async, batching, caching, error handling
 
         For changes to the encoding algorithm, risk thresholds, or scoring rules, please email
-        jeffmurr@seas.upenn.edu instead — those run server-side.
+        jeff.murray@alumni.upenn.edu instead — those run server-side.
 
   - type: dropdown
     id: kind

semanticembed-0.7.5/.github/dependabot.yml

@@ -0,0 +1,21 @@
+# Weekly update PRs for the pinned GitHub Actions and the Python dependencies.
+version: 2
+updates:
+  - package-ecosystem: github-actions
+    directory: /
+    schedule:
+      interval: weekly
+    groups:
+      actions:
+        patterns: ["*"]
+  - package-ecosystem: pip
+    directory: /
+    schedule:
+      interval: weekly
+    groups:
+      python:
+        patterns: ["*"]
+  - package-ecosystem: pip
+    directory: /huggingface
+    schedule:
+      interval: weekly

{semanticembed-0.7.3 → semanticembed-0.7.5}/.github/workflows/ci.yml

@@ -6,6 +6,9 @@ on:
   pull_request:
     branches: [main]
 
+permissions:
+  contents: read
+
 jobs:
   test:
     name: Test (${{ matrix.python-version }} on ${{ matrix.os }})
@@ -23,10 +26,12 @@ jobs:
             python-version: "3.12"
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          persist-credentials: false
 
       - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
         with:
           python-version: ${{ matrix.python-version }}
           cache: pip
@@ -43,9 +48,11 @@ jobs:
     name: Build wheel + sdist
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          persist-credentials: false
 
-      - uses: actions/setup-python@v5
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
         with:
           python-version: "3.12"
 

semanticembed-0.7.5/.github/workflows/corpus.yml

@@ -0,0 +1,42 @@
+name: Corpus
+
+# Manual only. Cloning the pinned public repos is not part of every PR.
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  score:
+    name: Score pinned corpus
+    runs-on: ubuntu-latest
+    environment: jev  # only jobs in this environment can read TYPESAFE_API_KEY
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          persist-credentials: false
+
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Install
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e '.[dev,extract]'
+
+      - name: Score tuning and held-out extraction
+        run: python scripts/corpus_score.py --split all
+
+      # Report only: Jev judges every extracted edge, including edges that
+      # appear only at each repo's current HEAD. Skipped without the secret.
+      - name: Jev edge audit
+        env:
+          TYPESAFE_API_KEY: ${{ secrets.TYPESAFE_API_KEY }}
+        run: |
+          if [ -z "$TYPESAFE_API_KEY" ]; then
+            echo "::notice::TYPESAFE_API_KEY is not set; the Jev edge audit was not run."
+            exit 0
+          fi
+          python scripts/jev/edge_audit.py --log edge_audit_log.jsonl

semanticembed-0.7.5/.github/workflows/jev-drift.yml

@@ -0,0 +1,40 @@
+name: Jev drift
+
+# Weekly check that the Jev model behind the leak guard still answers the
+# fixed cases in scripts/jev/canary_cases.json the way it did when the
+# guard was measured (scripts/jev/canary_baseline.json). Also reports, without
+# failing, how the moving alias jev-latest answers, so a new model release
+# shows up before the pinned version is changed.
+#
+# Needs the secret TYPESAFE_API_KEY in the `jev` environment. About 48 requests a run.
+
+on:
+  schedule:
+    - cron: "17 6 * * 1"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  drift:
+    name: Compare with baseline
+    runs-on: ubuntu-latest
+    environment: jev  # only jobs in this environment can read TYPESAFE_API_KEY
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          persist-credentials: false
+
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Install
+        run: python -m pip install "httpx==0.28.1"
+
+      - name: Run the drift check
+        env:
+          TYPESAFE_API_KEY: ${{ secrets.TYPESAFE_API_KEY }}
+        run: python -I scripts/jev/drift_canary.py --also jev-latest

semanticembed-0.7.5/.github/workflows/leak-guard.yml

@@ -0,0 +1,88 @@
+name: Leak guard
+
+# Checks the lines a pull request adds for internals of the hosted service:
+# a risk flag's cutoff, the steps that compute a value, how the input graph
+# is changed before scoring, or how ties are broken. Uses TypeSafe's Jev
+# model (scripts/jev/leak_guard.py, measurements in docs/jev.md).
+#
+# Needs the secret TYPESAFE_API_KEY in the `jev` environment. Without it the check fails
+# rather than passing unchecked. Pull requests from forks get no secrets, so
+# they need a maintainer's review for this instead.
+#
+# The guard's code is taken from the base branch, never from the pull
+# request, so a pull request cannot change what runs with the key.
+
+on:
+  pull_request:
+    branches: [main]
+
+permissions:
+  contents: read
+
+concurrency:
+  group: leak-guard-${{ github.event.pull_request.number }}
+  cancel-in-progress: true
+
+jobs:
+  leak-guard:
+    name: Check added lines
+    runs-on: ubuntu-latest
+    environment: jev  # only jobs in this environment can read TYPESAFE_API_KEY
+    timeout-minutes: 15
+    steps:
+      - name: Check out the pull request (data only)
+        uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+          fetch-depth: 0
+          path: pr
+          persist-credentials: false
+
+      - name: Check out the guard from the base branch
+        uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          ref: ${{ github.event.pull_request.base.sha }}
+          path: trusted
+          persist-credentials: false
+
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Install the guard's one dependency
+        run: python -m pip install "httpx==0.28.1"
+
+      - name: Run the leak guard
+        env:
+          TYPESAFE_API_KEY: ${{ secrets.TYPESAFE_API_KEY }}
+          BASE_SHA: ${{ github.event.pull_request.base.sha }}
+          HEAD_SHA: ${{ github.event.pull_request.head.sha }}
+          FROM_FORK: ${{ github.event.pull_request.head.repo.full_name != github.repository }}
+          ACTOR: ${{ github.event.pull_request.user.login }}
+        run: |
+          if [ ! -f trusted/scripts/jev/leak_guard.py ]; then
+            echo "::notice::The leak guard is not on the base branch yet. It checks pull requests once it is merged."
+            exit 0
+          fi
+          if [ -z "$TYPESAFE_API_KEY" ] && [ "$FROM_FORK" = "true" ]; then
+            echo "::error::Pull requests from forks get no secrets, so the leak guard cannot run. A maintainer must review the added lines."
+            exit 2
+          fi
+          if [ -z "$TYPESAFE_API_KEY" ] && [ "$ACTOR" = "dependabot[bot]" ]; then
+            echo "::notice::Dependabot pull requests get no repository secrets. Leak guard not run; review the version bumps by hand."
+            exit 0
+          fi
+          cd trusted
+          python -I scripts/jev/leak_guard.py --repo ../pr --base "$BASE_SHA" --head "$HEAD_SHA" \
+            --log ../leak_guard_log.jsonl --json-out ../leak_guard_results.json
+
+      - name: Keep the requests and answers
+        if: always()
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: leak-guard-log
+          path: |
+            leak_guard_log.jsonl
+            leak_guard_results.json
+          if-no-files-found: ignore
+          retention-days: 30

semanticembed-0.7.5/.github/workflows/publish.yml

@@ -0,0 +1,112 @@
+name: Publish to PyPI
+
+# Runs on a tag push of the form `v*`: git tag v0.7.5 && git push origin v0.7.5
+# Uses PyPI Trusted Publishing (OIDC), so no PyPI token is stored in GitHub.
+#
+# A release is published only when all of these hold:
+#   1. the tagged commit is on main (a tag on an unreviewed branch is refused);
+#   2. the tag matches pyproject.toml and semanticembed.__version__;
+#   3. the full test suite passes on that commit;
+#   4. someone approves the `pypi` environment. Set required reviewers in
+#      Settings -> Environments -> pypi, and restrict it to `v*` tags.
+#
+# One-time PyPI setup: pypi.org -> Account -> Publishing -> trusted publisher
+#   Project: semanticembed  Owner: jmurray10  Repository: semanticembed-sdk
+#   Workflow: publish.yml   Environment: pypi
+#
+# Actions are pinned to commit SHAs; Dependabot proposes updates.
+
+on:
+  push:
+    tags:
+      - "v*"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: publish
+  cancel-in-progress: false
+
+jobs:
+  verify:
+    name: Verify tag and run tests
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Tagged commit is on main
+        run: |
+          git fetch --no-tags origin main
+          if ! git merge-base --is-ancestor "$GITHUB_SHA" origin/main; then
+            echo "::error::Tag $GITHUB_REF_NAME points at a commit that is not on main"
+            exit 1
+          fi
+
+      - name: Tag matches the package version
+        run: |
+          python - <<'EOF'
+          import os, pathlib, re, sys, tomllib
+          tag = os.environ["GITHUB_REF_NAME"].removeprefix("v")
+          pkg = tomllib.loads(pathlib.Path("pyproject.toml").read_text())["project"]["version"]
+          init = re.search(r"__version__\s*=\s*[\"']([^\"']+)",
+                           pathlib.Path("src/semanticembed/__init__.py").read_text()).group(1)
+          if not tag == pkg == init:
+              sys.exit(f"::error::tag {tag}, pyproject.toml {pkg} and __version__ {init} must match")
+          print(f"version {tag}")
+          EOF
+
+      - name: Install
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e '.[dev,extract]'
+
+      - name: Run pytest
+        run: pytest -q
+
+  build:
+    name: Build sdist + wheel
+    needs: verify
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          persist-credentials: false
+
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Build
+        run: |
+          python -m pip install --upgrade build twine
+          python -m build
+          python -m twine check dist/*
+
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
+        with:
+          name: dist
+          path: dist/
+          if-no-files-found: error
+
+  publish:
+    name: Publish
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write   # trusted publishing; nothing else in this job can write
+    steps:
+      - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        with:
+          name: dist
+          path: dist/
+
+      - uses: pypa/gh-action-pypi-publish@7245906fb61fa33a13554f9d5fb856fe61758086 # v1.14.3

{semanticembed-0.7.3 → semanticembed-0.7.5}/CHANGELOG.md

@@ -2,6 +2,84 @@
 
 All notable changes to the SemanticEmbed SDK.
 
+## 0.7.5 — Security fixes, parser corrections
+
+### Security
+
+- **Agent no longer runs the scanned project's hooks.** `semanticembed-agent`
+  started the Claude Code CLI in your working directory with default
+  settings, so a cloned repo's `.claude/settings.json` hooks ran shell
+  commands with no prompt. The agent now loads no settings files, runs the
+  CLI in an empty temporary directory, and has no built-in tools (shell,
+  file read/write, web fetch). Requires `claude-agent-sdk>=0.1.60`; older
+  versions ignore `setting_sources=[]` and are refused at startup.
+- **`extract.from_requirements`** follows `-r` only to files inside the
+  starting file's directory, after resolving symlinks. Lines of a file
+  outside it used to become node names sent to the API.
+- **`find_edges()` LLM fallback** skips `node_modules`, `.venv`, `vendor`
+  and similar directories, skips files that resolve outside the scanned
+  path, and replaces secret-looking values with `[REDACTED]` before
+  sending file text to Anthropic or Google. README "What gets sent where"
+  now covers `find_edges()` and `explain()` / `ask()`.
+- **Skills** no longer pre-approve shell commands.
+- **Releases** publish only from a tag on `main` that matches the package
+  version, after the test suite passes, with approval on the `pypi`
+  environment. Workflow actions are pinned to commit SHAs; Dependabot is on.
+
+### Changed
+
+- **`extract.from_crewai`** reads `@CrewBase` / `@agent` / `@task` / `@crew`
+  crews: agent-to-task edges, the sequential task chain, YAML and inline
+  `context` dependencies, and `manager_agent` fan-out. crewAI-examples
+  `stock_analysis` goes from 0 to 7 edges and `surprise_trip` from 0 to 5.
+- **`extract.from_langgraph`** reads `Command(goto=...)`,
+  `Command[Literal[...]]`, and `Send(...)`, and keeps a separate `START` /
+  `END` per `StateGraph`. A compiled subgraph mounted with `add_node`, or
+  invoked with `ainvoke` / `invoke` / `stream` / `astream`, is wired to the
+  caller. `open_deep_research` `deep_researcher.py` goes from 6 edges (three
+  graphs sharing one `START`) to 16.
+- **`extract.from_terraform`** labels a resource `type.name` when its short
+  name is shared (`aws_vpc.this` vs `aws_internet_gateway.this`). On
+  terraform-aws-vpc this is 55 edges / 35 nodes with a `this` hub, then
+  119 edges / 96 nodes with no bare `this` once module blocks are included.
+  Correct keys put real modules over the free-tier 50-node cap.
+- **`extract.from_directory`** takes `include` and `exclude` source labels.
+  The default still parses every recognized source, including GitHub Actions
+  and Python imports.
+- **Kubernetes names that collide** (`Service` and `Deployment` both called
+  `frontend`) are labeled `Kind/name`. Same-name pairs used to become
+  self-loops and disappear. `GoogleCloudPlatform/microservices-demo`
+  `c7cb6d5` `release/` is 28 edges after that, not 17.
+- **Pulumi / CDK kwargs** such as `vpc_security_group_ids=[group.id]` count
+  as a reference to `group`.
+- **Python `from ._native import`** is resolved against the importing package.
+- **Corpus check** (`corpus/manifest.json`, `scripts/corpus_score.py`):
+  held-out extraction precision 1.0 and recall 1.0 on 2026-11-04
+  (55/55 edges, 6 files). Tuning was 1.0 (59/59). The only pin behind
+  HEAD was this repo, and those labeled files did not change edges.
+  Hosted scoring notes are in `docs/scoring_gaps.md`. The 0.7.4 changelog line "vpc goes
+  from 2 to 56 edges" does not reproduce on vpc `b3abd6df`: the indexed-ref
+  parser before name disambiguation returned 55 edges. With module blocks
+  and brace-matched bodies, vpc `b3abd6df` is 119 edges / 96 nodes
+  (root `*.tf` files alone: 93 edges / 73 nodes). `pyproject.toml` and
+  `requirements.txt` are parsers. A `]` inside a requirement extra does
+  not end a multiline TOML array.
+
+## 0.7.4 — Terraform count/for_each + Kubernetes auto-detect fixes
+
+- **`extract.from_terraform`** now resolves `count` / `for_each`-indexed
+  references (`aws_subnet.this[0].id`, `aws_eip.nat[count.index].id`) — the
+  form real modules use almost universally. The ref regex previously matched
+  only `type.name.` and missed them. The note below this release said
+  that went from 2 to 56 edges; a remeasure of vpc `b3abd6df` is under
+  Unreleased (119 edges / 96 nodes with the current parser).
+- **`extract.from_directory`** now detects Kubernetes manifests by content
+  (`apiVersion` + `kind`) in any directory, not just conventional names like
+  `k8s/` / `manifests/`. Manifests under `release/` or `kubernetes-manifests/`
+  (e.g. GoogleCloudPlatform/microservices-demo) are now picked up.
+- Both surfaced by validating the parsers against real public repos; added
+  `tests/test_extract_regressions.py` to lock them in.
+
 ## 0.7.3 — Modern AutoGen + lenient edge parser
 
 - `from_autogen` now recognizes the autogen-agentchat 0.4+ patterns:

{semanticembed-0.7.3 → semanticembed-0.7.5}/CODE_OF_CONDUCT.md

@@ -30,7 +30,7 @@ Examples of unacceptable behavior:
 ## Enforcement
 
 Instances of unacceptable behavior may be reported to the maintainer at
-**jeffmurr@seas.upenn.edu**. All complaints will be reviewed and
+**jeff.murray@alumni.upenn.edu**. All complaints will be reviewed and
 investigated promptly and fairly.
 
 The maintainer is obligated to respect the privacy and security of the

{semanticembed-0.7.3 → semanticembed-0.7.5}/CONTRIBUTING.md

@@ -67,4 +67,4 @@ git push origin vX.Y.Z
 
 ## Questions
 
-Open a GitHub Discussion or email jeffmurr@seas.upenn.edu.
+Open a GitHub Discussion or email jeff.murray@alumni.upenn.edu.