スキル一覧に戻る

redact-for-public-disclosure

pjt222
更新日 2 days ago
5 閲覧
17
2
17
GitHubで表示
コミュニケーションgeneral

について

このスキルは、リバースエンジニアリングの調査結果を公開前に体系的に秘匿情報を編集する手法を提供し、方法論と教育的価値を保持します。拒否リストパターン、Git履歴漏洩を防ぐオーファンコミット公開、未編集マージをブロックするCIゲートなどの技術を含みます。サードパーティツールの研究成果を公開する際や、機密情報を保護しながらアップストリーム提案を準備する場合にご利用ください。

クイックインストール

Claude Code

推奨
メイン
npx skills add pjt222/agent-almanac -a claude-code
プラグインコマンド代替
/plugin add https://github.com/pjt222/agent-almanac
Git クローン代替
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/redact-for-public-disclosure

このコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします

ドキュメント

Redact for Public Disclosure

Split reverse-engineering research repo into private source-of-truth and public-disclosure subset using redaction checker, pattern deny-lists, orphan-commit publish pattern. Methodology travels; specific findings stay private.

When Use

  • Publishing methodology findings about closed-source CLI harness you integrate with
  • Preparing upstream proposal or bug report to project you don't own
  • Archive private research repo as public reference
  • Promote investigation notes (Phase 1-4 artifacts) into public guide
  • Establish publish pipeline before findings pile up so leak risk no back up
  • Clean up after near-miss where draft almost shipped sensitive identifier

Inputs

  • Required: Private research repo with mixed-sensitivity content (source of truth)
  • Required: Target public mirror (separate repo, or public/ worktree) where redacted content publishes
  • Optional: Existing draft slated for publication
  • Optional: Version-lag policy (default: "current + 1 prior stays private")
  • Optional: List of vendor identifiers, flag prefixes, namespaces already known sensitive

Steps

Step 1: Categorize Every Candidate Fact

Before write or promote any content, sort each fact into one of four categories. Category determines whether and when it ships.

CategoryDefinitionShareable?
methodologyThe how of investigation, independent of any specific findingAlways
generic patternClass-level observations (e.g., "harnesses commonly use a single-prefix flag namespace")Yes
version-specific findingConcrete observation tied to a specific release (e.g., "in vN.M, the gate defaults off")Only after the version-lag cool-off
live internalMinified names, byte offsets, dark flag names, current-version gate logic, PRNG/salt constants, internal codenamesNever

Tag each draft section, capture log, or note with category before review for publication. Section that mixes categories splits — methodology lifts out clean. Rest stays private.

Got: Every candidate fact has category label. Drafts intended for public mirror contain only methodology and generic-pattern entries (plus version-specific findings older than cool-off).

If fail: Fact resists categorization? Treat as live internal by default. Re-categorize only after explicit review against version-lag policy.

Step 2: Set Version-Lag Cool-Off Policy

Decide up front how many versions sit between "current" and "shareable." Two is typical: current + 1 prior stay private; older patterns may be discussed. Write policy into private repo (e.g., REDACTION_POLICY.md) so future-you no need re-derive it.

# Redaction Policy

Version-lag cool-off: **2 releases**.
- Current release (vN): all version-specific findings PRIVATE.
- Previous release (vN-1): all version-specific findings PRIVATE.
- Releases vN-2 and earlier: version-specific findings may move to public draft after Step 5 review.

Source of truth for "current": output of `monitor-binary-version-baselines`.
Owner: <name>. Reviewed quarterly.

"Current" version must be empirical (read from installed binary), not administrative. Tie policy to baseline scanner output, not calendar.

Got: Committed REDACTION_POLICY.md in private repo with explicit cool-off and owner.

If fail: Stakeholders cannot agree on cool-off? Default to most conservative proposal. Cool-offs can shorten later; recall a leak cannot.

Step 3: Build Deny-List Scanner

Maintain patterns in single executable script that is source of truth for redaction policy. Script lives in private repo (tools/check-redaction.sh). Runs against public mirror.

#!/usr/bin/env bash
set -u
PUBLIC_REPO="${1:-./public}"
LEAKS=0

PATTERNS=(
  "minified identifier shape|<regex matching short bundle-style identifiers>"
  "vendor-prefixed flag|<regex matching the vendor's flag prefix>"
  "PRNG/salt constant|<regex matching the specific constants>"
)

for entry in "${PATTERNS[@]}"; do
  desc="${entry%%|*}"
  pattern="${entry##*|}"
  if rg -q "$pattern" "$PUBLIC_REPO"; then
    echo "LEAK: $desc"; LEAKS=$((LEAKS+1))
  fi
done
exit $LEAKS

Each entry has human-readable label and regex. One entry per sensitive identifier shape (not per literal string — shapes survive version churn). Exit code = number of leaks; clean run exits 0.

Got: tools/check-redaction.sh ./public-mirror runs in under a second on small repo. Exits 0 when nothing matches.

If fail: rg unavailable? Fall back to grep -rqE. Patterns too broad (every run reports leaks)? Narrow at source, no add suppressions.

Step 4: Maintain Deny-List Before Drafting

When Phase 1-4 finding could leak through draft, extend scanner before draft is written. Drafts cheap; teaching scanner new patterns durable.

Workflow:

  1. New finding lands in private repo (e.g., newly-discovered flag prefix).
  2. Ask: "If this leaked, what would I want the scanner to catch?"
  3. Add pattern entry to tools/check-redaction.sh (label + regex).
  4. Run scanner against entire public mirror to confirm new pattern not already tripped by legitimate content.
  5. Only then draft any public content that touches area.

This inverts usual order: scanner updates first, draft second. Scanner becomes executable specification of "what is too sensitive to publish." Draft cannot accidentally outpace it.

Got: Pattern entries in tools/check-redaction.sh predate any public-mirror content that could match them. git log tools/check-redaction.sh shows scanner updates landing before related draft commits.

If fail: Scanner updates lag drafts? Audit public mirror against new pattern immediately. Redact, then commit scanner update with note explaining discovered pattern.

Step 5: Establish Private/Public File-Set Split

Define explicit allow-list of files that sync to public mirror. New files default private; promotion requires redaction-check clearance.

# tools/public-allowlist.txt
README.md
LICENSE
guides/methodology-overview.md
guides/category-classification.md
docs/contributing.md

tools/sync-to-public.sh reads allow-list, copies only those files to public mirror, exits non-zero if allow-list references file that does not exist (catches typos).

#!/usr/bin/env bash
set -eu
PRIVATE_ROOT="${1:?private repo path required}"
PUBLIC_ROOT="${2:?public mirror path required}"
ALLOWLIST="$PRIVATE_ROOT/tools/public-allowlist.txt"

while IFS= read -r path; do
  [ -z "$path" ] && continue
  case "$path" in \#*) continue ;; esac
  src="$PRIVATE_ROOT/$path"
  dst="$PUBLIC_ROOT/$path"
  if [ ! -e "$src" ]; then
    echo "MISSING: $path"; exit 2
  fi
  mkdir -p "$(dirname "$dst")"
  cp -a "$src" "$dst"
done < "$ALLOWLIST"

Promotion needs three things in order: file added to allow-list, file passes redaction check, reviewer confirms category labels from Step 1.

Got: Public mirror contains exactly files listed in tools/public-allowlist.txt. No file appears in public mirror that is not on allow-list.

If fail: File appears in public mirror but missing from allow-list? Treat as leak event — investigate how it arrived, then either remove or formally promote it after redaction review.

Step 6: Publish via Orphan Commit

Public mirror is single git commit --orphan-rooted commit recreated at each publish. This prevents git log on public repo from exposing pre-redaction drafts.

# In the public mirror (separate repo or worktree)
cd /path/to/public-mirror
git checkout --orphan publish-tmp
git rm -rf .                                    # Clear the index
# Sync from private using the allow-list
bash /path/to/private/tools/sync-to-public.sh /path/to/private .
git add -A
git commit -m "Publish: <date>"
git branch -D main 2>/dev/null || true
git branch -m main
git push --force origin main

Public repo git log shows exactly one commit. Prior drafts and any redaction iterations stay in private repo history. No git log -p, git reflog, or branch listing on public repo can recover pre-redaction content because never committed there.

Got: git log --oneline on public mirror shows single commit per publish. No references to private repo history (no parent SHAs, no merge commits, no tags from private repo) appear.

If fail: git push --force rejected (branch protection)? Open single-commit pull request from clean orphan branch instead. Never solve rejection by pushing private history.

Step 7: Wire CI Gate

Run tools/check-redaction.sh on every commit to public-sync branch. Failed check blocks publish, not just warns.

# .github/workflows/redaction-check.yml (in the public mirror repo)
name: redaction-check
on:
  push:
    branches: [main, publish-*]
  pull_request:
    branches: [main]
jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Install ripgrep
        run: sudo apt-get update && sudo apt-get install -y ripgrep
      - name: Fetch redaction scanner
        env:
          GH_TOKEN: ${{ secrets.PRIVATE_REPO_TOKEN }}
        run: |
          gh api repos/<org>/<private-repo>/contents/tools/check-redaction.sh \
            --jq .content | base64 -d > check-redaction.sh
          chmod +x check-redaction.sh
      - name: Run scanner
        run: ./check-redaction.sh .

Two design choices:

  • Scanner pulled from private repo at CI time so deny-list itself never lives in public repo (patterns themselves sensitive — publishing them tells reader exactly what to look for).
  • Job exits with scanner exit code; non-zero blocks workflow.

Got: Pushes that introduce deny-listed pattern fail CI; publish does not land. Maintainers see failing label (e.g., LEAK: vendor-prefixed flag) without seeing regex itself.

If fail: Private-repo token cannot be granted to public CI? Embed only minimum-leak portion of scanner in public repo (broad shape patterns that no themselves identify vendor) and run full scanner pre-push from private repo.

Step 8: Handle False Positives Honestly

When scanner trips on legitimate content, prefer narrow pattern over add ignore-line. Broad deny-lists with local suppressions rot fast — six months later no one remembers why a particular line was suppressed, and next leak slides past unnoticed.

Decision tree:

  1. Is match actually safe? Re-categorize using Step 1. Content turns out to be live internal in disguise? Redact it; no suppress scanner.
  2. Is pattern too broad? Tighten regex so safe content no longer matches. Document tightening with comment in check-redaction.sh linking to case that motivated it.
  3. Only if 1 and 2 both fail — and pattern structurally too entangled with legitimate content to narrow further — use single-line suppression with # REASON: comment that states why suppression safe. Date the comment.
# Bad — mystery suppression
echo "API endpoint pattern" >> ignore.txt

# Good — narrowed pattern with rationale
# Pattern v2: tightened from `\bgate\(` to `\bgate\(['\"][a-z]+_phase` after
# legitimate `gate(true)` calls in our own SDK examples started matching. 2026-04-15.
PATTERNS+=("vendor flag predicate|\\bgate\\(['\"][a-z]+_phase")

Got: Each scanner pattern has zero or one inline comment explaining tightening. Suppressions, if any, carry date and rationale.

If fail: Suppressions accumulate (more than one per quarter)? Deny-list is mis-shaped. Schedule redaction-policy review. Rebuild patterns from categorized fact inventory.

Step 9: Periodic Redaction Sweeps

Not all redaction work is incident-driven. Run periodic sweep (monthly typical) that re-categorizes most recent additions to private repo and re-runs scanner against public mirror. Drift catches itself before it becomes incident-grade.

Sweep checklist:

  • Re-read version-lag policy; confirm empirical "current" version unchanged or update policy
  • Audit last month of private-repo commits for newly-added findings not categorized (Step 1)
  • Run tools/check-redaction.sh against public mirror (should still exit 0)
  • Review any scanner patterns added since last sweep — any too broad? Tighten if so
  • If any version aged past cool-off, ID findings now eligible for promotion
  • Confirm tools/public-allowlist.txt matches actual public-mirror file set

Got: Short sweep log per month in private repo (e.g., sweeps/2026-04.md) with checklist outcomes and any actions taken.

If fail: Sweep repeatedly skipped? Automate calendar reminder. Sweep keeps finding same drift? Workflow upstream is the problem — investigate why categorization gets skipped at draft time.

Checks

  • Every file in public mirror is on tools/public-allowlist.txt
  • tools/check-redaction.sh ./public-mirror exits 0
  • git log --oneline on public mirror shows single orphan commit per publish
  • REDACTION_POLICY.md exists in private repo with explicit version-lag cool-off
  • Every Phase 1-4 finding has category label (methodology / generic pattern / version-specific / live internal)
  • Public CI runs scanner on every push; deliberate test pattern fails build
  • Deny-list scanner itself does not live in public repo
  • Most recent monthly sweep log dated within last 35 days

Pitfalls

  • "Just one example to make it concrete." Temptation to include one specific finding "to ground methodology" = most common leak path. Use synthetic placeholders (e.g., acme_widget_v3, widget_handler_42) — clearly invented, never traceable to real product.
  • Use git rebase or git filter-branch to scrub leak in place on public repo. Force-pushing rewritten history still leaves traces in clones and forks. Orphan-commit publish pattern = structural fix; ad-hoc history rewriting = not.
  • Suppressions instead of pattern tightening. Scanner with twenty suppressions = scanner with zero meaningful coverage. Every suppression = future leak waiting for context to fade.
  • Public CI that warns instead of failing. Warnings get ignored. CI gate must block publish (non-zero exit, no merge button).
  • Allow-list drift. New files added to private repo do not automatically belong on allow-list. Default-deny = only safe posture.
  • Mistake encryption for redaction. Encoding, hashing, or rot13-ing sensitive identifier and publishing result still publishes it — original recoverable. Redact = "does not appear at all."
  • Publish the deny-list. Patterns themselves are finding catalog: reader who sees regex knows exactly what to grep for in binary. Keep scanner private; only its labels (e.g., LEAK: vendor-prefixed flag) should appear in public CI logs.
  • Treat private repo as draft pile. It is source of truth for research, not scratch space. Apply same versioning, review, backup discipline you would to any production artifact.

See Also

  • monitor-binary-version-baselines — Phase 1, baselines feed version-lag policy: what counts as "current" is empirical fact, not calendar fact
  • probe-feature-flag-state — Phases 2-3, classification findings here enter redaction pipeline at category step (Step 1)
  • conduct-empirical-wire-capture — Phase 4, capture artifacts (wire logs, payload schemas) need redaction before any can be referenced public
  • security-audit-codebase — both pipelines benefit from deny-list-style scanning; this skill specializes for research disclosure rather than secret leakage
  • manage-git-branches — orphan-commit publish pattern is branch operation; safe execution requires branch hygiene practices documented there

GitHub リポジトリ

pjt222/agent-almanac
パス: i18n/caveman/skills/redact-for-public-disclosure
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

関連スキル

himalaya-email-manager

コミュニケーション

このClaude Skillは、Himalaya CLIツールを介してIMAPを使用したメール管理を可能にします。開発者は自然言語クエリでIMAPアカウントからメールを検索、要約、削除できます。日次の要約取得やバッチ操作の実行など、Claudeから直接自動化されたメールワークフローにご利用いただけます。

スキルを見る

imsg

コミュニケーション

imsgは、macOS用のCLIツールで、Messages.appを介してプログラムでiMessage/SMSとやり取りできます。開発者がチャットの一覧表示、メッセージ履歴の閲覧、会話のリアルタイム監視、メッセージや添付ファイルの送信を可能にします。このスキルを使用して、メッセージングタスクを自動化したり、開発ワークフローにiMessage/SMS機能を統合したりできます。

スキルを見る

internationalization-i18n

コミュニケーション

このClaudeスキルは、アプリケーションの国際化(i18n)とローカライゼーション実装に関する包括的なガイダンスを提供します。i18nextやgettextなどのライブラリを使用したメッセージ抽出、翻訳管理、ロケール固有のフォーマット、RTL(右から左)対応などの主要タスクを網羅しています。多言語アプリケーションを構築する際や、国際ユーザー向けにローカライゼーション機能を追加する際にご活用ください。

スキルを見る

wacli

コミュニケーション

wacliは、WhatsApp Webプロトコルを介してWhatsAppメッセージの送信、検索、同期を可能にするコマンドラインツールです。主にClawdisワークフロー内で自動処理に使用されますが、メッセージ送信、チャット同期、履歴照会のために直接呼び出すこともできます。主な機能には、QRコード認証、バックグラウンドでの継続的同期、テキストおよびファイル送信機能が含まれます。

スキルを見る