Ghost Report Agent Skill

Combined Security Report

You aggregate findings from all scan skills (scan-deps, scan-secrets, scan-code) into a single prioritized report. Do all work yourself — do not spawn subagents or delegate.

$ARGUMENTS

Step 0: Setup

Run this Bash command to compute paths:

repo_name=$(basename "$(pwd)") && remote_url=$(git remote get-url origin 2>/dev/null || pwd) && short_hash=$(printf '%s' "$remote_url" | git hash-object --stdin | cut -c1-8) && repo_id="${repo_name}-${short_hash}" && short_sha=$(git rev-parse --short HEAD 2>/dev/null || date +%Y%m%d) && ghost_repo_dir="$HOME/.ghost/repos/${repo_id}" && scans_dir="${ghost_repo_dir}/scans/${short_sha}" && cache_dir="${ghost_repo_dir}/cache" && skill_dir=$(find . -path '*/skills/report/SKILL.md' 2>/dev/null | head -1 | xargs dirname) && echo "scans_dir=$scans_dir cache_dir=$cache_dir skill_dir=$skill_dir"

Store scans_dir (commit-level scan directory), cache_dir, and skill_dir.

Cache Check

If <scans_dir>/report.md already exists, show:

Combined security report is at: <scans_dir>/report.md

And stop. Do not regenerate it.

Step 1: Read Repo Context

Read <cache_dir>/repo.md if it exists. Extract:

• Business criticality
• Sensitive data types
• Component map

If it does not exist, continue without it — this is not an error.

Step 2: Discover Scan Results

List the contents of <scans_dir> to see which scan-type directories exist. Recognized types:

• deps/ — SCA / dependency vulnerability scan
• secrets/ — secrets and credentials scan
• code/ — code security scan (SAST)

If none of these directories exist, report an error:

No scan results found in <scans_dir>. Run one or more scan skills first:
  /ghost-scan-deps
  /ghost-scan-secrets
  /ghost-scan-code

And stop.

Step 3: Collect Findings

For each scan type that exists, glob <scans_dir>/<type>/findings/*.md and read each finding file in full. Retain the complete markdown body of every finding — the report will inline this content directly so readers never need to open individual finding files.

From each finding, also extract these metadata fields for filtering and sorting:

• ID — from ## Metadata → ID
• Type — the scan type (deps, secrets, or code)
• Severity — from ## Metadata → Severity (high, medium, low)
• Status — from ## Metadata → Status (e.g., confirmed-exploitable, unverified, verified, rejected, clean)

Step 4: Filter and Sort

Filter: Keep only high-confidence findings:

• For deps findings: status is confirmed-exploitable
• For secrets findings: status is NOT clean and NOT rejected
• For code findings: status is verified or unverified (NOT rejected)

Exclude any finding with status clean, rejected, or false-positive.

Sort the remaining findings:

• By severity: high first, then medium, then low
• Within same severity: deps before secrets before code

Step 5: Read Per-Scan Reports

For deps and secrets scan types, read <scans_dir>/<type>/report.md if present. Extract:

• Statistics (candidates scanned, confirmed findings, false positives filtered)
• Executive summary highlights

Note: code does not produce a report.md. For code scan coverage, count the finding files in <scans_dir>/code/findings/ directly. The "Candidates Scanned" count is the total number of finding files (all statuses). "Confirmed Findings" is the count with status verified, confirmed, or unverified. "False Positives Filtered" is the count with status rejected. Do NOT count clean file analyses from the nomination/analysis funnel — those never became findings.

If a per-scan report does not exist for deps or secrets, note it as unavailable.

Step 6: Generate Report

• Read <skill_dir>/report-template.md
• Populate the template with collected data:
• Fill Scan Information with repository name, commit SHA, date, and which scans ran
• Write Executive Summary using repo context and aggregated findings
• For all writing elements in this security-focused, objective and fact based report, use a neutral, human tone that balances expertise with ease of reading. Do not use emojis, em-dashes, etc.
• For Critical & High findings (severity = high): inline the substantive content from each finding file directly into the report — include code snippets, assessment tables, remediation commands, and all relevant detail so the report is fully self-contained
• For Medium findings: write a full subsection per finding with description, location, code context, and remediation (not a condensed table)
• Omit low-severity findings (they remain in per-scan finding files only)
• Fill Scan Coverage table from per-scan report statistics (for code, use finding file counts from Step 5)
• Add a brief methodology note per scan type that ran (1-2 sentences drawn from per-scan reports)
• Do NOT include links to per-scan reports or individual finding files — all content is inlined
• Write the report to <scans_dir>/report.md

Step 7: Show Output

Combined security report is at: <scans_dir>/report.md