Om Auto Sec Report

Auto Security Report — Driver

Aggregate a security analysis across a window of units of work. This skill does not perform the per-unit analysis itself; it delegates that to om-auto-sec-report-pr (the single-unit skill) and combines the resulting fragments into a single docs-only PR. The report preserves every per-unit finding, every "Apply elsewhere" pointer, and every "Next steps — go deeper" suggestion so the reviewer can keep drilling into specific areas with additional om-auto-sec-report-pr runs.

Arguments

{windowSpec} (optional) — one of:

• A date in YYYY-MM-DD form — every PR merged on or after that date into --base (default develop) up to today (UTC).
• A PR number — every PR whose number is greater than or equal to this value and that is merged into --base.
• A branch name (e.g. feat/foo) — treated as a single unit of work; the driver invokes auto-sec-report-pr branch:{name} exactly once and still produces the aggregate report layout.
• A spec path (any path ending in .md under .ai/specs/ or .ai/specs/enterprise/) — treated as a single unit of work; driver invokes auto-sec-report-pr spec:{path} exactly once.
• Omitted — defaults to the last 7 days (UTC) of merged PRs into --base.

Options:

• --base <branch> (optional) — base ref for PR / branch diffs. Defaults to develop. Merges into main are still reported and flagged when a different base is specified.
• --include-open (optional) — also include open non-draft PRs in the queue, flagged as "not yet merged". Off by default.
• --deep-scan (optional) — pass --deep-scan through to every sub-unit call so the apply-elsewhere sweeps cover the whole repo. Off by default.
• --max-units <n> (optional) — cap the number of sub-unit runs. Default: 50. Larger caps are allowed but keep the run paged.
• --slug <kebab-case> (optional) — override the slug used in the plan and artifact filenames. Default: derived from the window.
• --force (optional) — bypass the claim-conflict check when a previous run left a branch or plan behind.

Relationship to other skills

• Delegates to .ai/skills/om-auto-sec-report-pr/SKILL.md for every unit of work. All paranoid checks, deep vectors, apply-elsewhere sweeps, and next-step suggestions live in that skill. This driver only orchestrates.
• Reuses .agents/skills/om-auto-create-pr/SKILL.md for branch/worktree/commit/validation/label discipline when opening the aggregate PR.
• Hands off to .agents/skills/om-auto-continue-pr/SKILL.md when the run cannot finish in one invocation.

Workflow

0. Pre-flight and claim

Follow .agents/skills/om-auto-create-pr/SKILL.md step 0 verbatim.

DATE=$(date -u +%Y-%m-%d)
SLUG="${SLUG_OVERRIDE:-auto-sec-report-${DATE}}"
PLAN_PATH=".ai/runs/${DATE}-${SLUG}.md"
BRANCH="feat/${SLUG}"

1. Build the unit queue

Translate {windowSpec} into an ordered list of units. Each entry is one of pr:{n}, spec:{path}, or branch:{name}.

• Date mode: gh pr list --state merged --base {base} paginated until everything merged on or after the date is captured. Emit pr:{n} for each.
• PR number floor: same listing, filtered by number ≥ floor.
• Branch mode: a single-entry queue with branch:{name}.
• Spec mode: a single-entry queue with spec:{path}.
• Default: last 7 days of merged PRs.

Sort the queue newest-first for PR lists so the driver surfaces the most recent risk first. Honor --max-units; if the queue is longer, truncate with a noted residue in the plan's Risks section and propose a follow-up invocation for the remainder.

Record the resolved window (start date or PR floor, end date or target name, base branch, queue size) in the plan's Overview.

2. Draft the execution plan

Follow .agents/skills/om-auto-create-pr/SKILL.md step 3 with a Progress section shaped like this:

### Phase 1: Queue and plan

- [ ] 1.1 Resolve unit queue and record the window

### Phase 2: Per-unit analysis

- [ ] 2.N {target caption} — auto-sec-report-pr {target} --out-fragment ...

Phase 2 MUST list one checkbox per unit in the queue, in order. The checkbox title MUST include the exact om-auto-sec-report-pr command the driver will run. Append commit SHA on each flip so om-auto-continue-pr can resume precisely.

### Phase 3: Aggregation

- [ ] 3.1 Concatenate fragments into `.ai/analysis/auto-sec-report-${DATE}.md`
- [ ] 3.2 Build consolidated "Next steps — go deeper" list
- [ ] 3.3 Render HTML mirror to `.ai/analysis/auto-sec-report-${DATE}.html`
- [ ] 3.4 Spot-check artifacts: links resolve, redactions held, secret-grep clean

### Phase 4: PR delivery

- [ ] 4.1 Commit artifacts, push branch, open PR against `develop` (do not merge)
- [ ] 4.2 Apply `review`, `documentation`, `security`, `skip-qa` labels with comments

3. Isolated worktree and first commit

Follow .agents/skills/om-auto-create-pr/SKILL.md steps 4–5 verbatim.

4. Execute the queue

For each unit in Phase 2, invoke om-auto-sec-report-pr in sub-unit mode. Pass --out-fragment so no per-unit PR is opened and no autofix pass is triggered:

FRAGMENT_DIR=".ai/tmp/auto-sec-report/${DATE}-${SLUG}/fragments"
mkdir -p "$FRAGMENT_DIR"

for UNIT in "${QUEUE[@]}"; do
  FRAGMENT_PATH="${FRAGMENT_DIR}/$(slug "$UNIT").md"
  invoke_skill "om-auto-sec-report-pr" \
    --target "$UNIT" \
    --base "${BASE:-develop}" \
    ${DEEP_SCAN:+--deep-scan} \
    --out-fragment "$FRAGMENT_PATH"
done

Rules during the loop:

• Honor the in-progress lock protocol owned by om-auto-sec-report-pr when the unit is a PR. If a PR is locked by someone else, skip it, note the skip in the plan under the unit's Progress line, and continue.
• If a sub-unit exits non-zero, capture the partial fragment (if any), mark the unit's Progress line with ⚠ partial and the reason, and continue to the next unit. Do not abort the batch.
• Flip the unit's Progress checkbox to - [x] with the commit SHA once the fragment lands. Commit the progress update as its own docs(runs): mark ${SLUG} unit X complete commit.
• Push after every 5 units so om-auto-continue-pr always has a recent checkpoint to resume from.

5. Aggregate

After every unit is processed, build the aggregate markdown at .ai/analysis/auto-sec-report-${DATE}.md using this outline:

# Auto Security Report — {window caption}

Window: **{start date or PR floor} through {end date} | branch:{name} | spec:{path}** (base: `{base}`).
Units analyzed: {count} ({N PRs, M branches, L specs}).
Partial / skipped units: {if any, inline with reasons}.

## Executive Summary

- Total findings: {N blocker, M major, L minor, K nit, I info}.
- Top OWASP categories: {A01, A08, A10}.
- Top paranoid vectors surfaced across units: {TOCTOU, cache-key
  cross-tenant leakage, SSRF redirect chain, JWT alg confusion}.
- Single sentence on the riskiest residual area the reviewer should
  double-check.

## Consolidated Next Steps — Go Deeper

Every "Next steps" entry produced by per-unit fragments is listed
here, deduplicated and ordered by expected impact. The single
highest-impact entry is marked `[recommended]`.

- **[recommended]** `auto-sec-report-pr {target}` — {why}.
- `auto-sec-report-pr {target}` — {why}.
- Audit `packages/core/src/modules/sales/` for TOCTOU on concurrent
  shipment creation — {why}.
- ...

## Risk Heatmap

| OWASP Category | Blocker | Major | Minor | Notes |
|---|---|---|---|---|
| A01 Broken Access Control | {n} | {n} | {n} | {one sentence} |
| A02 Cryptographic Failures | {n} | {n} | {n} | {one sentence} |
| ... continue through A10 ... | | | | |
| Out of scope (not OWASP) | — | — | — | {n} findings |

## Paranoid Deep Vectors — Coverage Matrix

Row per vector, column per unit outcome (`covered`, `risk surfaced`,
`not applicable`, `inconclusive`). Abbreviated when units are many;
full tables remain in the per-unit fragments.

## Per-Unit Findings

{Concatenate every sub-unit fragment here, in queue order. Do not
rewrite the fragments — keep the per-unit "Next Steps" sections
intact so a reviewer can trace each consolidated entry back to its
unit.}

## Appendix — Queue

Each item below is one unit from the window with the exact invocation
that was run.

### {YYYY-MM-DD}

- `auto-sec-report-pr pr:{n}` — [#{n}]({url}) {title} — status: {complete|partial|skipped}
- `auto-sec-report-pr branch:{name}` — status: {...}
- ...

Rules for the aggregate:

• Do NOT paraphrase unit fragments. Include them verbatim under Per-Unit Findings.
• Deduplicate the consolidated "Next steps" list on exact command equality; keep the highest-severity justification.
• Mark exactly one [recommended] across the consolidated list, even when multiple units each marked their own.
• Never paste raw diffs, secrets, tokens, .env content, credentials, internal hostnames, or user PII. Redact to {REDACTED}.

6. Render HTML mirror

Write .ai/analysis/auto-sec-report-${DATE}.html following the HTML rules from .ai/skills/om-auto-sec-report-pr/SKILL.md step 6a:

• Stand-alone <!DOCTYPE html>, inline <style>, no JS, no remote assets, rel="noopener noreferrer" on every link.
• Mirror every section of the aggregate markdown.

7. Validation gate (docs-only)

Same as om-auto-sec-report-pr step 7:

• git diff --check on the artifact files.
• Secret-leak grep on the diff before commit.
• Manual re-read; every PR/issue/CVE link resolves.

8. Self-review and BC review

Apply .agents/skills/om-code-review/SKILL.md to the artifact diff. Verify no PII, no internal hostnames, no secrets leaked through.

9. Open the PR

Follow .agents/skills/om-auto-create-pr/SKILL.md step 9 with:

• Title: docs(analysis): add auto-sec-report for {window caption}.
• Base: develop. Never merge directly.
• Body MUST include Tracking plan: .ai/runs/${DATE}-${SLUG}.md and the correct Status: line.
• Body MUST link the aggregate markdown + HTML, state the queue size, the blocker/major count, and the top OWASP categories, and repeat the [recommended] next step verbatim so the reviewer can trigger the drill-deeper run in one line.

10. Labels

Apply in order, each with a short explanatory comment:

• review — "PR is ready for code review."
• documentation — "docs-only deliverable under .ai/analysis/."
• security — "security-posture report; merits a security-savvy reviewer."
• skip-qa — "docs-only report; no customer-facing behavior."

Never needs-qa.

11. Auto-review pass

Run .agents/skills/om-auto-review-pr/SKILL.md against the new PR in autofix mode. Apply fixes as new commits. Never rewrite history.

12. Summary comment

Post the comprehensive summary comment required by .agents/skills/om-auto-create-pr/SKILL.md step 12. In the "What can go wrong" section, be honest about report limitations:

• Findings are aggregated from per-unit heuristic analysis; confirm before acting.
• "Apply elsewhere" and "Next steps" pointers are suggestions; a human reviewer must confirm.
• A large window can mask per-unit context; if anything looks surprising, re-run the single-unit skill against the suspect unit.

13. Cleanup and resumability

Follow .agents/skills/om-auto-create-pr/SKILL.md step 13.

If the run cannot finish in a single invocation:

1. Leave Status: in-progress in the PR body.
2. Ensure the Phase-2 Progress checklist reflects which units completed and which are pending, each with commit SHAs and the exact om-auto-sec-report-pr command to resume.
3. Post a PR comment that says verbatim: 🤖 auto-sec-report is not complete. Resume with /auto-continue-pr {prNumber}.
4. Release any in-progress lock per om-auto-continue-pr rules.

Rules

• This skill delegates per-unit analysis to om-auto-sec-report-pr. Never re-implement the paranoid checks or the "Next steps" production here — always call the sub-unit skill.
• Always run in an isolated worktree. Never nest worktrees.
• Always open a docs-only PR against develop. Never merge from within this skill.
• Default window is the last 7 days of merged PRs (UTC) when {windowSpec} is omitted.
• Branch and spec inputs are first-class: a single-entry queue is still a valid driver run; the aggregate format is unchanged.
• Aggregate markdown concatenates sub-unit fragments verbatim. Do not rewrite them.
• Consolidated "Next steps" list MUST deduplicate on exact command equality and MUST mark exactly one entry [recommended].
• Never paste raw diffs, secrets, tokens, .env content, credentials, internal hostnames, or user PII.
• On partial batch runs (a unit skipped or failed), record the reason inline and continue — do not abort the whole driver.
• On partial completion of the driver itself, leave Status: in-progress and post a /om-auto-continue-pr {prNumber} hand-off comment.
• Labels: review, documentation, security, skip-qa. Never needs-qa.