Create force-multiplier delivery skill

[theMickster]

🎟️ Tracking

N/A

📔 Objective

Adds force-multiplier to bitwarden-delivery-tools — a skill for applying one intent across many targets as N consistent, idempotent, SSH-signed draft PRs across the Bitwarden GitHub enterprise.

The central design principle: the checks are the product. Making the edit is trivial. Making 74 edits provably correct, consistent, and reversible — and being able to prove what was done — is the actual hard part. The skill is built around that proof.

What ships:

• SKILL.md — a lean 71-line playbook built on a fixed pipeline: SELECT → CHECK YOURSELF → PILOT → FAN-OUT → REPORT → REMEDIATE. The skill never freestyles across the fleet; it compiles any user intent into a structured campaign spec and executes it deterministically.
• references/campaign-spec.md — generic campaign schema: intent + target selector + recipe + validation + PR spec + safety policy
• references/finding-targets.md — discovery library keyed by signal type (file-existence, content match, manifest inspection, topic/language filter), grounded in gh ... --owner bitwarden
• references/pipeline.md — per-stage mechanics: shallow-clone isolation, validation gates, idempotency rules, rate-limit handling, dry-run semantics
• references/safety-and-self-checks.md — the full checks-and-balances model: three reality-check brackets (before / on one / after), the mandatory pilot gate, the per-target second pass, the reference-check pre-step for destructive recipes
• examples/workflow-edit.md, examples/npm-to-pnpm.md, examples/settings-json-patch.md — fully-worked campaigns; named-change specifics live here, never in the skill body

The skill is deliberately generic — a compiled engine, not a catalogue. It composes with the existing delivery skills rather than reinventing them: perform-preflight → committing-changes → labeling-changes → creating-pull-request, per target, in that order.

Plugin README updated to include force-multiplier in the Mechanics skill table.

🧪 Testing

Executed end-to-end against a live AppSec remediation: VULN-649 — [VULN-649] ci: Remove deprecated scan workflow.

• 130 active repos enumerated; 74 confirmed to carry .github/workflows/scan.yml via per-repo contents API (not flaky code search)
• Pilot on bitwarden/admin: branch cut, git rm, full diff read line by line, PR spec locked before any fan-out
• Reference-check pre-step ran across the fleet before any removal (destructive recipe): branch-protection required-status-check query per repo; repos where protection could not be confirmed were held for AppSec/Platform — not auto-removed
• Fan-out in chunks of ≤10: per-target shallow clone, signal re-verified, idempotency gate, git rm, second-pass diff-shape check (exactly one deleted file — anything more stops that target), secrets-scan, commit, draft PR
• Commits: authored and signed as Mick Letofsky <mletofsky@bitwarden.com> via Secure Enclave, no Claude attribution

Total human input after confirming the pilot diff: one thumb on Touch ID per commit.

[github-actions]

Plugin Validation Summary — bitwarden-delivery-tools

Result: ✅ All checks passed. No critical, major, or blocking issues found. The PR adding the force-multiplier skill is structurally sound, version-consistent, secure, and ready to merge.

Validated the bitwarden-delivery-tools plugin (v2.1.0) and its new force-multiplier skill across three passes: plugin structure, skill quality, and security.

---

1. Plugin Validation (plugin-validator)

Status: PASS — 0 errors, 0 warnings.

Check	Result
plugin.json manifest (name, version, required fields, valid JSON)	✅ name: bitwarden-delivery-tools, version: 2.1.0 (valid semver), author/homepage/repository/keywords well-formed
Version consistency	✅ 2.1.0 agrees across plugin.json:3, .claude-plugin/marketplace.json entry, and CHANGELOG.md:8
Marketplace description parity	✅ marketplace description matches plugin.json
CHANGELOG (Keep a Changelog)	✅ New [2.1.0] - 2026-07-01 entry under ### Added documents both the skill and the eval set
Directory / auto-discovery	✅ Skill at skills/force-multiplier/SKILL.md; references/, examples/, evals/ present
Referenced-file integrity	✅ Every path cited in SKILL.md exists (all 5 references + 3 examples)
File organization	✅ Plugin README.md present and updated (README.md:41); no stray files (.DS_Store, logs, node_modules)
Hardcoded credentials	✅ None

Informational (not a defect): skills/force-multiplier/evals/evals.json line 35 (eval id 3) self-reports with-skill 6/9 vs baseline 7/9 on the "no-false-negatives / SELECT" expectation, flagged by the authors as a candidate for a stronger SELECT steer. Honest self-reporting, not a validation failure.

---

2. Skill Review (skill-reviewer) — force-multiplier/SKILL.md

Status: PASS — 0 critical, 0 major, 2 minor (both optional).

• Frontmatter: All fields valid — name matches directory; description, when_to_use, argument-hint, allowed-tools all present and well-formed. allowed-tools correctly declares the Skill(...) teaming dependencies used in the body.
• Description quality: Third-person/infinitive form, concrete, ~180 chars. Strong, specific trigger phrasings live in when_to_use ("across all repos", "fleet-wide", "mass update", etc.).
• Content: 1,244 words (within the 1,000–3,000 target); consistently imperative/infinitive writing; clear logical flow.
• Progressive disclosure: Exemplary. Lean core; detailed mechanics pushed to references/pipeline.md, schema to references/campaign-spec.md, safety to references/safety-and-self-checks.md, deferred scope to references/deferred.md, discovery to references/finding-targets.md; worked instances in examples/. All referenced files exist and are non-trivial.

Minor / warnings (should-fix, non-blocking):

• SKILL.md:11 — examples/ is referenced only as a directory, never by individual filename. Naming a representative example (e.g. npm-to-pnpm.md) would sharpen model discovery. Suggestion, not a fix.
• SKILL.md:4 — when_to_use is a deprecated field under current skill-authoring guidance (trigger phrasings are meant to fold into description); the field is still honored and the content is excellent. Low-priority stylistic call — consolidating would make description long.

Highlights: Strong safety framing (the "Check yourself, Claude" reality-check gate, non-negotiable safety defaults, treating target content as untrusted data for prompt-injection defense); reconciliation arithmetic that prevents silent target loss; clean teaming model resolving the interactive creating-pull-request dependency once at pilot.

---

3. Security Validation (reviewing-claude-config)

Status: PASS — no security issues.

• Committed secrets / hardcoded credentials: ✅ None. A full-plugin sweep for GitHub/Slack/AWS tokens and PEM private keys returned no matches. Every secret/token/credential hit is documentation of the skill's own secrets-scanning safety controls (e.g. safety-and-self-checks.md:63-71, SKILL.md:69-71). The AKIA[0-9A-Z]{16} string at safety-and-self-checks.md:65 is part of a recommended secrets-detection grep pattern, not a credential.
• Settings / permission scoping: ✅ No settings.json/settings.local.json/hooks files changed; nothing to scope.
• Dangerous auto-approvals / broad file access: ✅ None. allowed-tools (Bash, Read, Write, Edit, Glob, Grep, Agent, Skill(...)) is appropriate for a fan-out skill and scoped to the declared teaming skills.
• Security posture of the skill itself is a positive: treats all target-system content as untrusted data (CWE-1427 prompt-injection defense), enforces least privilege, per-target isolation, secrets-scan before every commit, draft-PRs-only, never force-push or auto-merge, and reuses sanctioned gh auth rather than injecting credentials.

---

Verdict

No must-fix errors. Two optional stylistic suggestions on SKILL.md (name a representative example; when_to_use deprecation). The force-multiplier addition meets all structural, versioning, changelog, referenced-file, skill-quality, and security requirements. Approved.

_{Note: scripts/validate-plugin-structure.sh and scripts/validate-marketplace.sh could not be executed in this sandbox (require approval); their checks were performed manually via file inspection and confirmed passing.}

[withinfocus]

❓ Did you use https://bitwarden.atlassian.net/browse/PM-35701?focusedCommentId=157547 at all? That was my attempt and helping show a prompt.

[theMickster]

No, I did not. I pointed Claude at https://pnpm.io/ website and let it go. I had Claude put a note here in the example. Perhaps in a follow-up commit I'll have it go through the commment e2e and see what comes of it.