Security Tools
Overview
This skill consolidates tools and workflows for analyzing the security posture, operational errors, and ownership metrics of an application.
Capabilities/Tools
Interact with the Sentry API to inspect application errors and tracebacks.
# Requires SENTRY_AUTH_TOKEN in the environment
python scripts/sentry_api.py --project my-project list-issues
SENTRY_BASE_URL must be HTTPS unless it is loopback. The client rejects
redirects and ambient proxies, accepts only the configured host, bounds every
response, and reads optional CA trust from SSL_CERT_FILE, SSL_CERT_DIR, or
REQUESTS_CA_BUNDLE without persisting environment-specific paths.
2. Security Ownership Maps (scripts/)
Analyze git histories and identify which users own the most sensitive files, or calculate bus factor (how catastrophic losing a single maintainer would be to security knowledge).
scripts/build_ownership_map.py: Parses git logs to output ownership data.
scripts/query_ownership.py: Interrogates the built ownership data.
scripts/community_maintainers.py: Resolves email boundaries for authors.
3. Threat Modeling & Best Practices (docs/)
Follow established application security frameworks from the docs/ directory when explicitly requested.
- Review design documents systematically using STRIDE.
- Apply secure-by-default coding conventions to Python, JS/TS, and Go codebases.
- Do NOT perform performative threat modeling unless explicitly requested by the user. Keep assessments focused on actual abuse paths, assets, and boundaries.
4. False-Positive Filtering (shared, CE-041)
When triaging a security register (your own findings, or code-enhancer's
analyze_security.py output), reduce noise with the shared two-stage filter
code-enhancer/scripts/findings_filter.py instead of hand-tuning regexes:
- Hard-exclusion stage (deterministic, no network) drops the chronic
low-signal classes — generic DOS/resource-exhaustion, "add rate limiting"
advice, resource leaks, open redirects, regex-injection, memory-safety findings
outside C/C++, and SSRF in static HTML. Findings on each finding's
{name, detail, file} shape.
- LLM confidence stage (opt-in) — in an agent-driven review (where a model
is available), pass a
judge(finding) -> (keep, confidence 1–10, reason)
callable to filter_findings(...); findings below confidence 8 are dropped,
and a judge error safely keeps the finding. Use this to apply
organization-specific context ("we use Cognito for auth; don't flag missing
password validation") — the natural-language equivalent of custom exclusion
rules.
from findings_filter import filter_findings # code-enhancer/scripts on path
kept, excluded, stats = filter_findings(register) # deterministic
kept, excluded, stats = filter_findings(register, judge=my_llm_judge) # + LLM
Borrowed from Anthropic's claude-code-security-review (MIT).
Related skill
1---2name: security-tools-23domain: infrastructure4skill_type: skill5description: >-6 Security analysis tools including checking Sentry error logs, conducting threat7 models, analyzing source code git repositories for security ownership and bus8 factors, and applying security best practices.9license: MIT10tags: [security, appsec, threat-model, ownership, sentry, errors, analysis]11metadata:12 source: https://skillmd.com/skills/knuckles-team/security-tools-213 version: '1.6.0'14 author: Genius15---1617# Security Tools1819## Overview2021This skill consolidates tools and workflows for analyzing the security posture, operational errors, and ownership metrics of an application.2223## Capabilities/Tools24Interact with the Sentry API to inspect application errors and tracebacks.25```bash26# Requires SENTRY_AUTH_TOKEN in the environment27python scripts/sentry_api.py --project my-project list-issues28```2930`SENTRY_BASE_URL` must be HTTPS unless it is loopback. The client rejects31redirects and ambient proxies, accepts only the configured host, bounds every32response, and reads optional CA trust from `SSL_CERT_FILE`, `SSL_CERT_DIR`, or33`REQUESTS_CA_BUNDLE` without persisting environment-specific paths.3435## 2. Security Ownership Maps (`scripts/`)36Analyze git histories and identify which users own the most sensitive files, or calculate bus factor (how catastrophic losing a single maintainer would be to security knowledge).37- `scripts/build_ownership_map.py`: Parses git logs to output ownership data.38- `scripts/query_ownership.py`: Interrogates the built ownership data.39- `scripts/community_maintainers.py`: Resolves email boundaries for authors.4041## 3. Threat Modeling & Best Practices (`docs/`)42Follow established application security frameworks from the `docs/` directory when explicitly requested.43- Review design documents systematically using STRIDE.44- Apply secure-by-default coding conventions to Python, JS/TS, and Go codebases.45- Do NOT perform performative threat modeling unless explicitly requested by the user. Keep assessments focused on actual abuse paths, assets, and boundaries.4647## 4. False-Positive Filtering (shared, CE-041)48When triaging a security register (your own findings, or `code-enhancer`'s49`analyze_security.py` output), reduce noise with the shared two-stage filter50`code-enhancer/scripts/findings_filter.py` instead of hand-tuning regexes:5152- **Hard-exclusion stage** (deterministic, no network) drops the chronic53 low-signal classes — generic DOS/resource-exhaustion, "add rate limiting"54 advice, resource leaks, open redirects, regex-injection, memory-safety findings55 outside C/C++, and SSRF in static HTML. Findings on each finding's56 `{name, detail, file}` shape.57- **LLM confidence stage** *(opt-in)* — in an agent-driven review (where a model58 is available), pass a `judge(finding) -> (keep, confidence 1–10, reason)`59 callable to `filter_findings(...)`; findings below confidence 8 are dropped,60 and a judge error safely **keeps** the finding. Use this to apply61 organization-specific context ("we use Cognito for auth; don't flag missing62 password validation") — the natural-language equivalent of custom exclusion63 rules.6465```python66from findings_filter import filter_findings # code-enhancer/scripts on path67kept, excluded, stats = filter_findings(register) # deterministic68kept, excluded, stats = filter_findings(register, judge=my_llm_judge) # + LLM69```7071Borrowed from Anthropic's `claude-code-security-review` (MIT).7273## Related skill7475- [security-tools-security-threat-model](../security-tools-security-threat-model/SKILL.md) — the security-threat-model sub-skill, served as its own skill.