Docs workflow security review

Skill opendatahub-io/docs-skills/skills/docs-workflow-security-review

Security and PII scan of documentation drafts before publication. Runs the deterministic PII scanner, then applies the docs-review-security checklist for agent-based analysis. Iteration logic is owned by the orchestrator, not this skill.From its SKILL.md

Install

npx -y skills add opendatahub-io/docs-skills --skill docs-workflow-security-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.5 KB, as published. Nobody here has run it

Security Review Step

Step skill for the docs-orchestrator pipeline. Follows the step skill contract: parse args → run scanner → apply checklist → write output.

Arguments

$1 — JIRA ticket ID (required)
--base-path <path> — Base output path (e.g., .agent_workspace/proj-123)

Input

<base-path>/writing/

Output

<base-path>/security-review/review.md
<base-path>/security-review/scanner-results.json
<base-path>/security-review/step-result.json

Execution

1. Parse arguments

Extract the ticket ID and --base-path from the args string.

Set the paths:

OUTPUT_DIR="${BASE_PATH}/security-review"
OUTPUT_FILE="${OUTPUT_DIR}/review.md"
SCANNER_FILE="${OUTPUT_DIR}/scanner-results.json"
mkdir -p "$OUTPUT_DIR"

2. Determine source files

Read the writing step's sidecar at ${BASE_PATH}/writing/step-result.json to determine the writing mode and file list.

If the sidecar exists and mode is "update-in-place" with a non-empty files array:

Build a file list from the files array.

Otherwise (draft mode, missing sidecar, or empty files array):

Set DRAFTS_DIR="${BASE_PATH}/writing" and glob for .adoc, .md, .dita, and .ditamap files.

3. Run PII scanner

python3 ${CLAUDE_PLUGIN_ROOT}/skills/docs-review-security/scripts/pii_scanner.py scan <file1> <file2> ... > "$SCANNER_FILE"
SCANNER_EXIT=$?

Check the exit code. The scanner uses these exit codes:

0 — no critical findings (warnings such as URL patterns may still exist in the JSON output)
1 — critical findings detected (credentials, private keys — proceed but flag in report)
2 — script error (bad paths, unreadable files, invalid arguments)

if [ $SCANNER_EXIT -gt 1 ]; then
  echo "ERROR: PII scanner failed (exit $SCANNER_EXIT). See output above." >&2
  exit 1
fi

Do NOT treat exit codes 0 or 1 as script failures — both produce valid JSON output. Exit 0 means no critical findings (but warnings may exist). Exit 1 means critical findings were detected.

Validate the JSON output is well-formed before parsing:

jq empty "$SCANNER_FILE" 2>/dev/null
if [ $? -ne 0 ]; then
  echo "ERROR: Scanner produced invalid JSON output." >&2
  exit 1
fi

Always parse the JSON to get finding counts — do NOT infer counts from the exit code. Read summary.total_findings, summary.by_severity.critical (default 0), and summary.by_severity.warning (default 0) from $SCANNER_FILE. These are the values to use in the report header.

4. Build report header

Start the review report with the scanner results:

# Security and PII Review — <TICKET>

## Automated scan results

**Scanner findings:** N total (C critical, W warnings)

[If findings > 0, list them grouped by severity]

## Agent analysis

[Apply the checklist from step 5 and add findings here]

5. Dispatch the security-reviewer agent (Layer 2)

You MUST use the Agent tool to invoke the security-reviewer subagent. Do NOT read the checklist or apply it yourself — the agent reads the source files and the Layer 2 checklist in its own isolated context and appends findings directly to the report, so neither the doc content nor the checklist enters the orchestrator's context.

Agent tool parameters:

subagent_type: docs-skills:security-reviewer
description: Security Layer 2 review for <TICKET>
run_in_background: false (the orchestrator must wait for the reviewer to finish before verifying output)

Prompt (substitute <SOURCE_FILES> with the file list from step 2 and <OUTPUT_FILE> with the report path):

Apply the Layer 2 agent-analysis checklist to the documentation for ticket <TICKET>.

Source files — review each of these: <SOURCE_FILES>

Report file: <OUTPUT_FILE> — this file already contains the report header and scanner results. Append your findings to its Agent analysis section by editing it in place. Do NOT overwrite the existing content and do NOT write to any other location.

After appending all findings, do NOT print the report contents. Print ONLY these two lines:
Written <OUTPUT_FILE>
Agent findings: N

6. Verify output

After the agent completes, verify the review report exists at <OUTPUT_FILE>.

7. Write step-result.json

Do not hand-author the sidecar — a hand-written sidecar drifts from the schema. Run the script, which derives every scanner field from $SCANNER_FILE and sums context_size_bytes from the output folder. Pass --agent-findings from the Agent findings: N line the security-reviewer agent printed (do not read the full report back to recount):

python3 ${CLAUDE_SKILL_DIR}/scripts/write_step_result.py \
  --ticket "<TICKET>" \
  --scanner-results "$SCANNER_FILE" \
  --agent-findings <N> \
  --output-dir "$OUTPUT_DIR" \
  --sidecar "${OUTPUT_DIR}/step-result.json"

The script writes the conformant step-result.json with scanner_findings, critical_findings, agent_findings, per-category counts, context_size_bytes, and a real wall-clock completed_at. If the script exits non-zero, the sidecar was not written — fix the scanner-results path and re-run; do not substitute a stub.