protectai / llm-guard
The Security Toolkit for LLM Interactions
protectai/llm-guard holds a health index of 19 out of 100, placing it in the Critical band. It scores highest on Engineering Quality (84/100) and lowest on Vitality (32/100). The repository is archived, so no further maintenance is expected.
Software health index
Metrics are grouped into weighted categories on one standardized 1–100 scale. Overall starts as their weighted mean, calibrated against the distribution of the public record so bands carry percentile meaning; when public evidence triggers the High-Risk Jurisdiction Policy, the rating is adjusted and receives an At Risk ceiling of 34.
Score profile
Each axis is a category. The shape matters more than the average — a healthy subject fills the whole shape, while a spike-and-crater profile means strength in one dimension is masking risk in another.
The weighted overall 61 is calibrated to 67 on the published index scale (record calibration 2026-08-02).
Ownership
This repository is backed by an organization — shared, accountable stewardship that can outlive any single maintainer.
Package ecosystems
| Registry | Package | Version | Downloads / mo | Versions | Last publish | Tags |
|---|---|---|---|---|---|---|
| PyPI | llm-guard | 0.3.16 | 121,760 | 26 | 486 days ago | llmlanguage-modelsecurityadversarial-attacksprompt-injectionprompt-leakagepii-detectionself-hardeningfirewall |
Metrics by category
Vitality
Is the project alive — is code being written and are releases shipping?
Development activity
Work has arrived and nobody has attended to it
Silence alone is never the finding — a small, finished library may need no commits for years. This alert requires both: no human commit for a long time, and obligations the project is visibly not discharging. Automated commits do not count as maintenance.
How abandonment is assessedHow it's scored
| 18/36 | Push recency — last push 70 days ago |
| 0.7/36 | Commit cadence — 1/52 weeks with commits |
| 2.7/18 | Commit volume — 1 commits in the last year |
| 0/10 | OpenSSF Scorecard: Maintained — project is archived |
Inputs used
| commits_last_year | 1 |
| human_commit_share | 0.72 |
| days_since_last_push | 70 |
| active_weeks_last_year | 1 |
Release discipline
How it's scored
| 16.2/27 | Ships releases — 23 version tags (no GitHub releases) |
| 7.2/36 | Release recency — latest release 486 days ago |
| 19.8/27 | Release cadence — a release every ~60.1 days |
| 0/10 | OpenSSF Scorecard: Signed-Releases — no data |
Inputs used
| releases_count | 23 |
| latest_release_tag | v0.3.16 |
| releases_from_tags | yes |
| days_since_latest_release | 486 |
| mean_days_between_releases | 60.1 |
Community & Adoption
Does the project have users, downloads, attention, and a welcoming setup for contributors?
Popularity & adoption
How it's scored
| 56.9/60 | Stars — 3,207 stars |
| 22.2/25 | Forks — 461 forks |
| 8/15 | Watchers — 29 watchers |
Inputs used
| forks | 461 |
| stars | 3,207 |
| watchers | 29 |
| growth_state | unverified |
| growth_factor_pct | 100 |
| growth_unverified_reason | no_history |
Community health
How it's scored
| 22.5/22.5 | README |
| 22.5/22.5 | License — recognized license (MIT) |
| 18/18 | CONTRIBUTING guide |
| 0/13.5 | Code of conduct |
| 0/7.2 | Issue template |
| 6.3/6.3 | PR template |
Inputs used
| has_readme | yes |
| has_license | yes |
| readme_badges | 4 |
| has_contributing | yes |
| has_issue_template | no |
| has_code_of_conduct | no |
| readme_badge_services | shields.io |
| has_pull_request_template | yes |
Ecosystem adoption (downloads)
How it's scored
| 67.8/80 | Monthly downloads — 121,760 downloads/month across pypi |
| 0/20 | Registry dependents — not reported by this ecosystem |
Inputs used
| packages | llm-guard |
| dependents | — |
| ecosystems | pypi |
| total_downloads | — |
| monthly_downloads | 121,760 |
| unverified_packages_excluded | — |
Sustainability & Governance
Will the project survive its people — bus factor, responsiveness, who backs it, and package upkeep?
Maintainer resilience (bus factor)
How it's scored
| 9/54 | Bus factor — 1 contributor(s) cover half of all commits |
| 1.6/22.5 | Commit distribution — top contributor authored 93% of commits |
| 13.5/13.5 | Contributor breadth — 20 contributors |
| 0/10 | OpenSSF Scorecard: Contributors — project has 0 contributing companies or organizations -- score normalized to 0 |
Inputs used
| bus_factor | 1 |
| contributors_sampled | 20 |
| top_contributor_share | 0.927 |
How it's scored
| 37.7/42 | Issue resolution — 90% of issues closed |
| 16.3/30 | PR acceptance — 113/208 decided PRs merged |
| 0/13 | Newcomer PR acceptance — 0/2 first-time contributors' PRs merged in 30d |
| 12/15 | OpenSSF Scorecard: Code-Review — Found 16/18 approved changesets -- score normalized to 8 |
Inputs used
| merged_prs | 113 |
| open_issues | 12 |
| closed_issues | 104 |
| prs_merged_7d | 0 |
| prs_decided_7d | 0 |
| prs_merged_30d | 0 |
| prs_decided_30d | 2 |
| issue_closed_ratio | 0.897 |
| closed_unmerged_prs | 95 |
| first_time_authors_30d | 1 |
| first_time_prs_merged_30d | 0 |
| first_time_prs_decided_30d | 2 |
Ownership & stewardship
How it's scored
| 30/30 | Ownership backing — organization-owned |
| 0/20 | Verified domain |
| 20.8/25 | Owner reach — 787 followers of protectai |
| 17.9/25 | Track record — 16 public repos, account ~4 yr old |
Inputs used
| followers | 787 |
| owner_type | Organization |
| is_verified | no |
| owner_login | protectai |
| public_repos | 16 |
| account_age_days | 1,627 |
Package maintenance
How it's scored
| 25/25 | Published & resolvable — 1 package(s) on pypi |
| 14/35 | Publish recency — latest publish 486 days ago |
| 20/20 | Version history — 26 published versions |
| 20/20 | Not deprecated — active, not deprecated or yanked |
Inputs used
| packages | llm-guard |
| ecosystems | pypi |
| any_deprecated | no |
| min_days_since_publish | 486 |
Engineering Quality
Are baseline engineering and documentation practices in place?
Engineering practices
How it's scored
| 24/24 | CI workflows — 4 workflow(s) |
| 24/24 | Tests present |
| 16/16 | Linter config — .ruff.toml |
| 9.6/9.6 | Pre-commit hooks |
| 6.4/6.4 | .editorconfig |
| 0/20 | OpenSSF Scorecard: CI-Tests — 2 out of 21 merged PRs checked by a CI test -- score normalized to 0 |
Inputs used
| has_ci | yes |
| has_tests | yes |
| has_editorconfig | yes |
| has_linter_config | yes |
| has_precommit_config | yes |
Documentation
How it's scored
| 30/30 | README |
| 25/25 | Documentation directory |
| 15/15 | Documentation / homepage site — https://protectai.github.io/llm-guard/ |
| 10/10 | Repository description |
| 10/10 | Topics — 10 topics |
| 0/10 | Wiki |
Inputs used
| topics | large-language-models, llm, chatgpt, security-tools, adversarial-machine-learning, prompt-engineering, prompt-injection, llmops, llm-security, transformers |
| has_wiki | no |
| homepage | https://protectai.github.io/llm-guard/ |
| docs_site | https://protectai.github.io/llm-guard/ |
| has_readme | yes |
| has_docs_dir | yes |
| has_description | yes |
Security
Are visible security and supply-chain practices strong, without unresolved high-risk jurisdiction exposure?
Security posture
How it's scored
| 7.5/7.5 | Binary-Artifacts — no binaries found in the repo |
| 0.8/7.5 | Branch-Protection — branch protection is not maximal on development and all release branches |
| 0/2.5 | CI-Tests — 2 out of 21 merged PRs checked by a CI test -- score normalized to 0 |
| 0/2.5 | CII-Best-Practices — no effort to earn an OpenSSF best practices badge detected |
| 6/7.5 | Code-Review — Found 16/18 approved changesets -- score normalized to 8 |
| 0/2.5 | Contributors — project has 0 contributing companies or organizations -- score normalized to 0 |
| 10/10 | Dangerous-Workflow — no dangerous workflow patterns detected |
| 7.5/7.5 | Dependency-Update-Tool — update tool detected |
| 0/5 | Fuzzing — project is not fuzzed |
| 2.5/2.5 | License — license file detected |
| 0/7.5 | Maintained — project is archived |
| 0/5 | Packaging — no data |
| 0/5 | Pinned-Dependencies — dependency not pinned by hash detected -- score normalized to 0 |
| 3.5/5 | SAST — SAST tool detected but not run on all commits |
| 5/5 | Security-Policy — security policy file detected |
| 0/7.5 | Signed-Releases — no data |
| 0/7.5 | Token-Permissions — detected GitHub workflow tokens with excessive permissions |
| 7.5/7.5 | Vulnerabilities — 0 existing vulnerabilities detected |
Inputs used
| source | openssf_scorecard |
| checks_evaluated | 16 |
| scorecard_version | v5.5.0 |
| checks_inconclusive | 2 |
| scorecard_aggregate | 5.4 |
Dependency advisories
How it's scored
| 6/35 | Direct dependencies free of known advisories — 3 affected: transformers 4.51.3 (critical 9.6), json-repair 0.44.1 (high 7.5), nltk 3.10.3 (high 7.0) |
| 8.6/25 | Indirect dependencies free of known advisories — 1 affected: cryptography 44.0.3 (high 8.2) |
| 27.7/40 | No advisories left outstanding — 2 advisory-carrying package(s) unaddressed past 90 days; oldest published 437 days ago |
Inputs used
| source | osv |
| advisories | 40 |
| affected_packages | 4 |
| assessed_packages | 97 |
| unassessed_packages | 0 |
| affected_by_severity | critical 1, high 3 |
| direct_affected_packages | 3 |
AI Readiness
How well is the repo equipped to be developed and maintained with AI coding agents? Carries a deliberately small weight (4%): agent tooling is a real maintenance signal, but a repository with none can still reach 100/100.
Agent context & guidance
How it's scored
| 0/45 | Agent instructions — no CLAUDE.md / AGENTS.md / editor rules |
| 0/15 | Machine-readable docs (llms.txt) |
| 18.5/40 | Legible commit history — 25 of 72 human commits state their intent (structured subject or explanatory body) |
Inputs used
| has_llms_txt | no |
| llms_txt_url | — |
| legible_history_share | 0.347 |
| agent_instruction_files | — |
| agent_instruction_max_bytes | — |
Verify loop (build / test / typecheck)
How it's scored
| 18/18 | One-command bootstrap — Makefile, llm_guard_api/Makefile |
| 22/22 | Automated tests |
| 11/11 | Lint / format config — .ruff.toml |
| 11/11 | Static type checking — pyrightconfig.json |
| 10/10 | Reproducible environment — Dockerfile |
| 2/10 | Demonstrated agent practice — 1 of the last 100 commits agent-authored or agent-credited |
| 8/8 | Automated maintenance — 28 of the last 100 commits are automated dependency updates |
| 0/10 | OpenSSF Scorecard: Pinned-Dependencies — dependency not pinned by hash detected -- score normalized to 0 |
Inputs used
| has_nix | no |
| has_tests | yes |
| lockfiles | — |
| has_dockerfile | yes |
| typed_language | no |
| bootstrap_files | Makefile, llm_guard_api/Makefile |
| has_devcontainer | no |
| has_linter_config | yes |
| typecheck_configs | pyrightconfig.json |
| agent_commit_share | 0.01 |
| toolchain_manifests | — |
| dependency_bot_commit_share | 0.28 |
Code legibility for models
How it's scored
| 27/45 | Type-checkable code — Python with type-check config (pyrightconfig.json) |
| 55/55 | Manageable file sizes — 0/217 source files over 60KB |
Inputs used
| primary_language | Python |
| largest_source_bytes | 26,353 |
| source_files_sampled | 217 |
| oversized_source_files | 0 |
Machine-readable interfaces
How it's scored
| 40/40 | API schema (OpenAPI/GraphQL/proto) — llm_guard_api/openapi.json |
| 0/20 | MCP server |
| 40/40 | Runnable examples — examples, notebooks |
Inputs used
| example_dirs | examples, notebooks |
| has_mcp_signal | no |
| api_schema_files | llm_guard_api/openapi.json |
| interfaces_expected_of | network-service |
Key facts
Data collection warnings
- Star history unavailable: GitHub GraphQL error: Resource not accessible by personal access token
- Could not fetch pypi package 'llm-guard-api' from its registry
More detail
Star and fork history 0 ★ / 461 ⇿
When each star and fork was added, collected from GitHub and bucketed by day. Cumulative growth sits directly above the daily additions it is made of, so the two read against each other: steady organic accretion looks nothing like an abrupt, short-lived burst. Where that difference is measurable, it is reported as growth authenticity.
Each point covers 3 days.
OpenSSF Scorecard 5.4 / 10
Independent, tool-agnostic security assessment from the open-source OpenSSF Scorecard. Each check rewards a security practice, not a specific vendor's tool. Checks Scorecard could not determine are marked n/a and excluded from the security score (never counted as zero).
| 10 | Binary-Artifacts | no binaries found in the repo |
| 1 | Branch-Protection | branch protection is not maximal on development and all release branches |
| 0 | CI-Tests | 2 out of 21 merged PRs checked by a CI test -- score normalized to 0 |
| 0 | CII-Best-Practices | no effort to earn an OpenSSF best practices badge detected |
| 8 | Code-Review | Found 16/18 approved changesets -- score normalized to 8 |
| 0 | Contributors | project has 0 contributing companies or organizations -- score normalized to 0 |
| 10 | Dangerous-Workflow | no dangerous workflow patterns detected |
| 10 | Dependency-Update-Tool | update tool detected |
| 0 | Fuzzing | project is not fuzzed |
| 10 | License | license file detected |
| 0 | Maintained | project is archived |
| n/a | Packaging | packaging workflow not detected |
| 0 | Pinned-Dependencies | dependency not pinned by hash detected -- score normalized to 0 |
| 7 | SAST | SAST tool detected but not run on all commits |
| 10 | Security-Policy | security policy file detected |
| n/a | Signed-Releases | no releases found |
| 0 | Token-Permissions | detected GitHub workflow tokens with excessive permissions |
| 10 | Vulnerabilities | 0 existing vulnerabilities detected |
Direct dependencies 28
| Registry | Package | Version constraint | Manifest |
|---|---|---|---|
| PyPI | asyncio | ==3.4.3 | llm_guard_api/pyproject.toml |
| PyPI | fastapi | ==0.115.12 | llm_guard_api/pyproject.toml |
| PyPI | llm-guard | ==0.3.16 | llm_guard_api/pyproject.toml |
| PyPI | pydantic | ==2.11.4 | llm_guard_api/pyproject.toml |
| PyPI | pyyaml | ==6.0.2 | llm_guard_api/pyproject.toml |
| PyPI | uvicorn | ==0.34.2 | llm_guard_api/pyproject.toml |
| PyPI | structlog | >=24 | llm_guard_api/pyproject.toml |
| PyPI | slowapi | ==0.1.9 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-instrumentation-fastapi | ==0.54b1 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-api | ==1.33.1 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-sdk | ==1.33.1 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-exporter-otlp-proto-http | ==1.33.1 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-exporter-prometheus | ==0.54b1 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-sdk-extension-aws | ==2.1.0 | llm_guard_api/pyproject.toml |
| PyPI | opentelemetry-propagator-aws-xray | ==1.0.2 | llm_guard_api/pyproject.toml |
| PyPI | psutil | >=5.9 | llm_guard_api/pyproject.toml |
| PyPI | bc-detect-secrets | ==1.5.43 | pyproject.toml |
| PyPI | faker | >=37,<38 | pyproject.toml |
| PyPI | fuzzysearch | >=0.7,<0.9 | pyproject.toml |
| PyPI | json-repair | ==0.44.1 | pyproject.toml |
| PyPI | nltk | >=3.9.1,<4 | pyproject.toml |
| PyPI | presidio-analyzer | ==2.2.358 | pyproject.toml |
| PyPI | presidio-anonymizer | ==2.2.358 | pyproject.toml |
| PyPI | regex | ==2024.11.6 | pyproject.toml |
| PyPI | tiktoken | >=0.9,<1.0 | pyproject.toml |
| PyPI | torch | >=2.4.0 | pyproject.toml |
| PyPI | transformers | ==4.51.3 | pyproject.toml |
| PyPI | structlog | >=24 | pyproject.toml |
All dependencies 0
Full resolved dependency set from the GitHub dependency graph: 0 direct and 0 indirect (transitive) packages. The transitive closure is complete when the repository commits a lockfile.
| Registry | Package | Version | Relation |
|---|
Dependency advisories 4
Installing pypi:llm-guard@0.3.16 pulls in 97 packages, direct and transitive: 4 carry known advisories, of which 3 are direct dependencies.
| Package | Version | Relation | Severity | Advisories | Fixed in |
|---|---|---|---|---|---|
| transformers | 4.51.3 | direct | critical | 28 | 5.10.0 |
| json-repair | 0.44.1 | direct | high | 1 | 0.60.1 |
| nltk | 3.10.3 | direct | high | 1 | — |
| cryptography | 44.0.3 | indirect | high | 10 | 50.0.0 |
An advisory means the version recorded in the dependency graph falls inside an advisory’s affected range. Reachability is not analysed, and the graph includes development and test pins — a finding may concern tooling rather than shipped software.
Raw JSON report machine-readable
Feedback
Spotted something off in this report, or have thoughts to share? Wrong measurements, missed tooling, ideas, questions — anything is welcome. Every message is read and gets a response.
Scores are signals, not warranties. They reflect publicly visible practices on GitHub — not a code audit, and not a security guarantee.
Missing data is excluded and weights renormalized, never scored as zero. Methodology is versioned and open: metrics v2.10.0, schema v0.34.0 — full methodology · metrics wiki.
How one result sits in the wider record: aggregate statistics — PyPI.