Claude Skills by bordenet

Blast radius analysis - search for ALL usages before modifying any existing code. Prevents breaking unrelated consumers by scoping impact before scoping fix.

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

Trusted-advisor gate for branch and PR hygiene. Auto-invokes when the user mentions creating a branch, opening a PR, shipping/promoting/deploying, cherry-picking, rebasing, hotfixing, or back-syncing. Suggests and explains in-script (preflight always exits 0) and writes a .branch-flow-cleared sentinel that pre-push Gate 3 hard-consumes on pushes to dev/main. Three escape hatches: per-branch ack file (touch .git/base-advisory-ack-<branch>), GIT_BASE_OVERRIDE=1 env var, or use an exempt prefix ...

Pull gate — MANDATORY before any work on an existing shared branch. git fetch + status check before touching code, running tests, or making changes. Fires whenever resuming, continuing, or reviewing work on a branch that exists on a remote.

Use when reviewing code changes — one combined reviewer by default, parallel specialized reviewers only on diff signals. Invoke as: /sp-cr-battery [min-score] [--security|--no-security] [--mode=bug-fix|feature] (optional 1.0–10.0 quality threshold, default 7.0; default 9.2 in Bug Fix Review Mode). Bug Fix Mode auto-activates on hotfix/* and fix/[A-Z]+-[0-9]+ branches.

Use when investigating a feature area, doing an RCA, or auditing code across multiple repos and perspectives. Runs structured searches through 8 lenses (security, auth, data flow, error handling, config/secrets, test coverage, API surface, cross-repo consistency) and produces a findings table with evidence.

Audit CODEOWNERS for drift: finds unowned files, dead rules (patterns matching zero files), and invalid owner references. Works with GitHub and GitLab remotes, and degrades to advisory-only output when neither CLI is authenticated. Run when CODEOWNERS is out of date, files lack coverage, or ownership is unclear.

Use when Biome reports noExcessiveCognitiveComplexity, when functions have deeply nested conditionals (3+ levels), or when refactoring large functions.

Compare significant design approaches before implementation: generate three real options, use a compact matrix and separated hostile review, test edge cases, and converge in two to three rounds. Not for brainstorming, implementation, or code review.

PREVIEW - Conductor-led bounded investigation for complex distributed system incidents. Serial or parallel branches. Produces incident packets. Persistence tooling not yet implemented.

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

Use when building, deploying, and documenting skills from a completed domain-design output. Handles the walking skeleton -> remaining P0s -> cookbook -> handoff cycle.

Synthesizes evidence from all investigator branches into a root cause verdict. Builds reasoning trees, detects contradictions, weighs evidence strength over agent count, and produces a ranked diagnosis. Dispatched by debug-conductor.

Use when executing an approved written plan natively in the current session with required verification and review

Use when reviewing, auditing, or fixing any script that invokes an external CLI (claude, gh, kubectl, aws, npm, terraform, helm, etc.). Forces behavioral grounding — enumerate the CLI's flags and defaults via `<cli> <subcommand> --help` BEFORE declaring the script correct. Catches silent defaults that bash syntax checks cannot.

DEFAULT WORKFLOW for ANY code change. Orchestrates the full rigorous development lifecycle: brainstorming → think-twice → debate → progressive-harsh-review → plan-and-execute → progressive-harsh-review → commit. This fires AUTOMATICALLY for code changes unless the user explicitly opts out. Sequences existing skills so no phase is skipped.

Use when renaming fields, changing API contracts, or refactoring data models across multiple services. Prevents incomplete dependency analysis — the #1 source of production incidents from "complete" work. Requires tracing READ → STORE → PASS paths.

Use when implementation is complete, all tests pass, and you need to decide how to integrate the work. Mandates autonomous code review (Step 0) before presenting options. Guides completion of development work by presenting structured options for merge, PR, or cleanup.

Use when running git checkout -b, git switch -c, git worktree add -b, or any command that creates a new work branch — enforces semantic prefix naming.

Use glab for GitLab merge requests, CI/CD, variables, issues, and raw API calls on gitlab.com or any self-hosted GitLab instance. Includes macOS/Windows install and auth walkthrough. Triggers on glab, GitLab CLI, GitLab MR, pipeline status, install/setup glab, or GitLab-specific repo operations.

Gates `hotfix/*` branches (and `fix/*` when `CHARTER_BRANCH_PREFIXES` is set) on a HOTFIX-CHARTER.md (symptom + diff budget + cr-battery pre-commit verdict + refactoring disposition). Prevents the incident-2026-1507 failure mode (a hotfix that grew to 73 files / +8,750 / -4,195 LOC because nobody asked 'is this the smallest thing that fixes the symptom?').

Use when implementing large issues across multiple sessions. Creates and maintains a living progress document that tracks completed work, decisions, refinements, and findings.

Review a proposed incident revert for the protection it removes, the failure it may restore, and the adjacent path to check. Produce a read-only risk note before the change decision.

Specialized investigator for diagnosing infrastructure, configuration, and deployment failures: config changes, resource exhaustion, deployment regressions, cloud provider issues, and environment mismatches. Dispatched by debug-conductor.

Persists debugging investigation context (hypotheses, evidence, eliminated approaches) across sessions. Companion to systematic-debugging. Use when starting, resuming, or handing off a multi-turn debugging session.

Partition any target skill.md into a resident safety kernel and an on-demand reference. Proposes a split using keyword scoring, applies on confirmation, and installs a permanent context-budget regression test.

Specialized investigator for diagnosing LLM/prompt behavior issues: tool selection failures, prompt regressions, context window problems, and parsing failures. Dispatched by debug-conductor as part of forked debugging.

Primary, default reviewer for ANY skill.md file or skill-adjacent tooling infrastructure -- covers both LLM-execution safety (determinism, shell portability, tool contracts, cross-agent compatibility) and prose/design quality (correctness, simplicity, blind spots, verifiability, operational risk, folded in from progressive-harsh-review) in one pass. Reviews superpower skills, shell tooling, installers, MCP/tool contracts, and agent-specific integration files as infrastructure consumed by fron...

Hard stop before gh pr merge, glab mr merge, or any forge REST merge endpoint. These call the forge API directly -- git hooks and pre-push gates never fire. Requires an explicit human merge utterance in the current conversation, distinct from PR/MR web-UI approval and distinct from push approval.

Per-batch adversarial review for ANY code change. 3 critic personas score on 5 dimensions each. Score <8 average = REJECT + rework. Faster than full review, more rigorous than lint.

Hard gate. Use BEFORE describing, summarizing, or approving any generated output — files, PDFs, API responses, script results. Fires on the ACTION PATTERN of generating output then describing it. Prevents confabulation disguised as verification. If there is no tool call (view, read, grep, open) between generate and describe, the description is fiction. Fires BEFORE verification-before-completion.

Semantic redundancy gate -- before debugging CI or fixing any pull request, verify the PR's stated goals are not already on the target branch. Close obsolete PRs immediately rather than waste tokens on CI triage.

Pre-commit quality gate - run lint, typecheck, test LOCALLY before committing. Prevents wasted CI time and embarrassing build failures.

Use when: committing or pushing code changes. Mandatory progressive review loop via sub-agent-code-reviewer. Skip only when the user explicitly says to skip review.

Adversarial review for plans, specs, and other non-code deliverables. One persona for reversible work, three for irreversible (ADRs, contracts, migrations). Route code to code-review-battery and skill files to llm-skill-review.

Code review gate - apply engineering rigor when reviewing PRs. Trace data flow, check blast radius, verify integration points.

Hard stop before any git push to a remote. Requires explicit human approval in the current conversation -- push triggers CI/CD, overwrites remote history with --force, and is not reliably reversible. Covers the discipline the pre-tool-use-red-autonomy.sh hook enforces mechanically; adds the sub-agent rules and the what-does-NOT-count table.

Use when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation

Specialized investigator for testing hypotheses through reproduction attempts. Designs experiments, executes controlled tests, and reports whether a hypothesis can be confirmed or rejected. Dispatched by debug-conductor.

Use when presenting code changes to a human for review — dispatches code-review-battery. Self-fires on intent to present (see Cardinal Rule in skill body). Skips battery dispatch if valid sentinel exists for clean HEAD. Always runs battery before presenting, never after.

Validate product / PM requirements before design or roadmap commitment — customer-grounded, stakeholder-owned, business-measurable. USE INSTEAD OF requirements-validation when author is a PM.

Use when validating feature requirements before design or implementation. Tests each requirement for falsifiability, measurability, and independence. Detects contradictions and guides resolution without resolving silently.

Pre-push advisory gate that compares the branch's cumulative diff against the linked Linear ticket's point-estimate. Warns when LOC exceeds N times the estimate (default 2x at 200 LOC/point). Surfaces the incident-2026-1507 failure mode (1pt ticket -> +8,750 LOC) at push time without blocking by default.

Cold-start advisory: surfaces commits on remote-tracking refs that landed in the last 24 hours and are NOT yet on any local branch. Catches the incident-2026-1507 pattern where a sibling machine pushed work between sessions and the next session edited blind. Read-only; never blocks.

Use when reviewing skill organization, rebalancing domains, adjusting loading triggers, or responding to structural review signals. Triggers on "reorganize skills", "skill hierarchy needs adjustment", "too many skills in domain", "split this skill folder", "domain rebalancing".

Proactive adversarial bug hunt. A sub-agent collects candidates; each candidate flows through five named gates (anti-hallucination re-read, sibling-divergence, test-coverage adjudication, reachability evidence, confidence scoring) and a final verification pass. Outputs ranked Confirmed bugs plus two named risk lists (Unreachable, Low-Confidence) plus a transparent rejection log. Not for debugging a known failure (sp-debug) or secret scanning (sp-scan).

Specialized investigator for diagnosing state consistency failures: replication lag, cache staleness, event ordering issues, cross-service data divergence, and eventual consistency bugs. Dispatched by debug-conductor.

Use when executing implementation plans with independent tasks in the current session

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

Use when implementing any feature or bugfix, before writing implementation code