Cve Source Check Agent Skill

CVE Source Check

Audits CVE/vulnerability source coverage for a technology stack. Given an inventory

of components and (optionally) the feeds you currently monitor, it maps each

component to authoritative CVE sources, flags gaps, and emits audit-ready reports.

Scope

| In scope | Out of scope |

|---|---|

| Mapping components → authoritative feeds via a versioned registry | Running vulnerability scanners (Trivy/Snyk/etc.) |

| Reporting coverage and gaps in JSON + Markdown | Fetching CVE content or ranking by severity |

| Optional HEAD-check for source URL reachability | Integrating with private/commercial vuln databases |

| Audit-ready output (deterministic, reproducible) | Live LLM research per run |

Inputs

| Flag | Purpose |

|---|---|

| --inventory <file> | JSON inventory: [{name, version?, type?}, ...] or {components: [...]}. |

| --inline "name@ver,name,..." | Quick comma-separated list. Mutually exclusive with --inventory. |

| --current-sources <file> | Optional. One URL per line. Blank lines and # comments skipped. |

| --service <name> | Free-form name used in report header and filenames. |

| --check-urls | HEAD-check every source URL (5s timeout, graceful degradation). |

| --registry <path> | Override default tech-source-registry.json. |

| --out-dir <path> | Output directory (default: cwd). |

JSON inventory format only. YAML is not supported — stdlib does not ship a YAML parser.

Outputs

| File | Format |

|---|---|

| cve-source-report-{service}-{YYYYMMDD}.md | Human-readable audit report. |

| cve-source-report-{service}-{YYYYMMDD}.json | Machine-readable per references/output-formats.md. |

| Exit code | Meaning |

|---|---|

| 0 | Full coverage. |

| 1 | Gaps exist (unmapped components or unmonitored sources). |

| 2 | At least one source URL is unreachable (only with --check-urls). |

| 3 | Input error (missing/malformed registry or inventory). |

Workflow

Phase 1: LOAD

• Locate the registry: tech-source-registry.json next to this SKILL.md by default.
• Build an inventory:
• From --inventory: parse JSON; accept either a list or {components: [...]}.
• From --inline: split on commas, parse name@version pairs.
• If --current-sources is provided, read URLs (one per line); normalize for

case-insensitive comparison.

Gate: at least one inventory component is present. Empty inventory → exit 3.

Phase 2: MAP & VERIFY

• For each component, look up name (and aliases) in the registry.
• Found → status mapped, attach the registry's source list.
• Missing → status unmapped, sources [].
• If current sources were loaded, mark each source monitored: true when its

normalized URL appears in the set.

• If --check-urls is set, HEAD-check every unique source URL. Treat

200/301/302/403/405 as reachable; record definite failures and network errors

distinctly. See references/source-verification.md.

Gate: every component has a status; every source has monitored and

reachable fields populated (reachable: null when checks are skipped).

Phase 3: REPORT

• Compute the summary: components, mapped/unmapped, monitored, coverage %,

gaps, unreachable.

• Write the JSON report.
• Write the Markdown report:
• Summary table.
• Components table with ✅ / ⚠️ / ❌ markers.
• Gaps section listing primary then secondary sources to add (only when

gaps exist).

• Unmapped section listing registry-extension TODOs (only when unmapped

components exist).

• Print a one-screen summary to stdout including report paths.
• Set the exit code per the table above.

Gate: both files exist on disk and the summary printed; exit code reflects

the audit result.

Quick start

# Inline, offline, no monitoring data
python3 scripts/check-cve-sources.py \
  --inline "[email protected],[email protected],postgres@16,redis@7,[email protected]" \
  --service my-service

# Inventory file + current monitored feeds
python3 scripts/check-cve-sources.py \
  --inventory examples/inventory.example.json \
  --current-sources examples/current-sources.example.txt \
  --service my-service

# Same, with link verification
python3 scripts/check-cve-sources.py \
  --inventory examples/inventory.example.json \
  --current-sources examples/current-sources.example.txt \
  --service my-service \
  --check-urls

Extending the registry

To add a technology, edit tech-source-registry.json. Each entry needs name,

aliases, type, and 1–3 sources. Schema lives at

references/registry-schema.md.

Reference Loading Table

| Signal | Load These Files | Why |

|---|---|---|

| adding a technology to the registry | registry-schema.md | Defines registry shape and allowed values. |

| checking source URLs | source-verification.md | Defines HEAD-check semantics and graceful degradation. |

| generating audit reports | output-formats.md | Defines JSON and Markdown report contracts. |

Error handling

"ERROR: failed to load registry"

Cause: registry file missing or malformed JSON.

Solution: confirm tech-source-registry.json is at --registry (or default

location) and parses with python3 -m json.tool.

"ERROR: failed to load inventory"

Cause: inventory file missing, malformed JSON, or unexpected shape.

Solution: validate with python3 -m json.tool. Inventory must be a list or an

object with a components key.

"ERROR: inventory is empty"

Cause: no usable components after parsing.

Solution: confirm each entry has a name. Inline form requires non-empty tokens.

Coverage stuck at 0%

Cause: --current-sources URLs do not match registry URLs exactly (e.g., extra

path segments, trailing slashes).

Solution: copy URLs directly from the registry. The script normalizes scheme/host

case and trailing slash; everything else must match.

--check-urls flags many [—] entries

Cause: network issues (proxy, DNS, offline) — recorded as reachable: null.

Solution: re-run without --check-urls for the audit; investigate network

separately. Network errors do not affect the gap exit code.