tornhill
01Plugin Claude Code
Critique-led architecture views and a security/privacy/compliance/cost audit for Claude Code, grounded in real code and git history.
1 1mo ago A
tokens not measured
original Apache-2.0
tornhill
02Plugin Claude Code
Critique-led architecture views for Claude Code: a high-altitude, code- and git-grounded system map whose headline output is a findings overlay (hidden coupling, hotspots, SPOFs, performance/efficiency/correctness pains, edge-case risks). Ships a companion /audit skill for security, privacy/compliance, legal, and cost.
1 1mo ago A
tokens not measured
original Apache-2.0
audit
03Skill Claude CodeCodex
Review a codebase for SECURITY, PRIVACY/COMPLIANCE, LEGAL, and COST risk — and ground every finding in real code AND a real, named standard clause (GDPR, HIPAA/FHIR, PCI-DSS, ISO 20022/PSD2, OWASP, ISO 27001, CCPA, WCAG) selected for the app's domain and jurisdiction. Two modes: a quick repo-wide posture scan, and a…
1 1mo ago A 186 tokens
original Apache-2.0
tornhill
04Skill Claude CodeCodex
Generate an "astronaut's view that is still analyzable" of a project's architecture — a high-altitude system map whose HEADLINE output is a critique layer (single points of failure, hidden coupling, performance/efficiency/ correctness pains, edge-case risks, recurring pains) pinned onto a structure + flow substrate…
1 1mo ago A 172 tokens
original Apache-2.0
cold-run
05Plugin Claude Code
Plugin marketplace listing 1 plugin: cold-run.
0 9d ago A
tokens not measured
cold-run
06Plugin Claude Code
Cold-run your agent skills to find which ones change nothing, plus 24 skills that survived the same test.
0 9d ago A
tokens not measured
Command Claude Code
Vet a rule, principle, or heuristic with the Delta Test v2 and decide if it can become a real Claude skill or is a platitude that should die.
0 9d ago A 33 tokens
Command
Cold-run every skill in a directory against a baseline agent and report which ones change nothing, which make the output worse, and which earn their place.
0 9d ago A 0 tokens
vet-routing
09Command
Test whether each skill's description actually gets that skill loaded, by routing written requests against descriptions alone.
0 9d ago A 20 tokens
Skill Claude CodeCodex
When shipping a retry, backoff, cache, fallback, autoscaler, rate limit or CI rerun policy that many callers share: run it forward under a SUSTAINED outage, not a one-off blip, and add a stop-condition (breaker, jitter, cap, single-flight) if load multiplies. This is the pile-on question — not duplicate charges…
0 9d ago A 129 tokens
Skill Claude CodeCodex
When a test asserts on a mock, spy, call count or private member: for each assertion complete "if ONLY this fails, the caller sees ". No honest answer means replace it with a return-value or read-back-state check. Keep call assertions only for effects that leave the process (email, payment, published event). Triggers…
0 9d ago A 99 tokens
Skill Claude CodeCodex
Before scripting a manual workflow someone described: label each step ESSENTIAL, COMPENSATING (it only undoes an earlier defect) or VESTIGIAL, fix the upstream defect instead of encoding the repair, and turn what the human caught by eye into assertions. Triggers: "automate this", "every Monday we", "then we fix /…
0 9d ago A 83 tokens
Skill Claude CodeCodex
When choosing between two or more named architectures or datastores: tabulate each candidate against each constraint, label every constraint value MEASURED / STATED / GUESSED / ABSENT, and let only non-guessed constraints eliminate. Name which cell decided, and the threshold that reopens it. Skip when only one…
0 9d ago A 109 tokens
Skill Claude CodeCodex
When a retry or timeout goes inside a request handler that must answer within someone else's clock (gateway, client, parent RPC, a checkout or claims endpoint): compute attempts x per-attempt timeout + backoff, check it fits the remaining budget, and fail inside your own clock. Per-phase timeouts sum. This is the…
0 9d ago A 88 tokens
Skill Claude CodeCodex
Before saying work is complete, ready, or nearly done: mark each of error paths, edge cases, tests, migration, permissions, observability, config, real data size, docs, rollout, cleanup, CI as DONE / NOT-STARTED / NOT-NEEDED, and report the count. Skip on a typo fix. Triggers: "is it ready", "write the PR…
0 9d ago A 92 tokens
Skill Claude CodeCodex
Before deleting a guard, null check, retry, sleep, or workaround that looks redundant: recover why it exists via git blame -w -C and git log -S, and check the data. Delete only when the originating case is provably gone, citing it in the commit. Triggers: "remove the dead check", "this is unnecessary", "nothing uses…
0 9d ago A 81 tokens
grant-diff
17Skill Claude CodeCodex
When editing an existing IAM policy, role, scope set or permission grant: enumerate what the code actually calls, subtract it from what the policy allows, and delete the excess with each removal's failure symptom named. Skip a fresh tight policy. Triggers: "update the IAM policy", "add a permission", "it needs access…
0 9d ago A 72 tokens
Skill Claude CodeCodex
When a retried call can create or accumulate something (charge, insert, send, publish): name the key that makes attempt two return attempt one, name what enforces it, and drop the retry where no key exists. The key must be committed before the first attempt. This is the duplicate question only — not retry overload…
0 9d ago A 92 tokens
Skill Claude CodeCodex
When shipping a product change meant to move a behaviour or a number — a paywall, free trial, onboarding flow, pricing page change, upsell sheet, banner, prompt or nudge: state the hypothesis, the metric, the instrument (and whether it exists), the baseline read BEFORE merge, and a removal threshold with a date and an…
0 9d ago A 123 tokens
Skill Claude CodeCodex
When writing a load or performance test that will support a claim: vary identity, data size, cache state, endpoint mix, arrival pattern and write/read ratio instead of holding them constant, mark each production value SOURCED or ASSUMED, and never report the number without its shape. Skip smoke tests. Triggers: "load…
0 9d ago A 81 tokens
Skill Claude CodeCodex
Before executing a task with 5+ separable units: score each on value and cost, rank by leverage, name what you will DO, DEFER and DROP, and abort if the distribution is flat rather than forcing an 80/20 cut. Where the units are measurable, measure first. Triggers: open-ended cleanup, "speed this up", "improve X".
0 9d ago A 83 tokens
Skill Claude CodeCodex
When designing a function or API whose easiest call is destructive or lossy: make the easy path recoverable and put the irreversible operation behind a different name plus explicit confirmation. Add friction only to the one-way operation, never to the common one. Triggers: "add a delete/purge/reset function", "wipe"…
0 9d ago A 71 tokens
Skill Claude CodeCodex
When writing a multi-step implementation plan: list each step's falsifiable assumption, score how many steps it kills, and run the cheap high-kill checks BEFORE building — with the exact command, and no reporting of evidence you did not collect. Skip mechanical plans with nothing to falsify. Triggers: "write a plan"…
0 9d ago A 82 tokens
Skill Claude CodeCodex
When one slow dependency can exhaust a shared connection pool, thread pool or worker set: compute slots = peak rps x hold time per consumer, cap the optional consumers, and reject at the cap without waiting. A breaker will not fire when slow calls still succeed. Skip when every consumer is core. Triggers: "one slow…
0 9d ago A 76 tokens