← All creators

MrBinnacle

User

Claude Skills

73 indexed · 0 Featured · 0 stars · avg score 65
Prolific

Categories

Indexed Skills (73)

AI & Automation Listed

hidden-and-plugin-skill-reachability

Diagnose "that skill doesn't exist" when the skill is installed and working. Use when: (1) the Skill tool refuses a skill name the user just typed as `/name`, (2) a skill named in CLAUDE.md or a docs file is absent from the available-skills listing, (3) `find ~/.claude/plugins -maxdepth 4 -iname "*skill*"` returns nothing for a plugin skill you know is installed, (4) a skill works for the operator but not for the agent, (5) a skill is present in the plugin cache and still unreachable. Three independent mechanisms make a present skill unreachable — `disable-model-invocation: true` frontmatter, plugin-cache directory depth, and project-level `skillOverrides` — and each produces the same symptom. Includes the verification commands for all three and the rule for documenting a skill list so this cannot recur.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

agent-definition-snapshot-at-session-start

A subagent reports a tool as unavailable and the harness says it is "disabled for this session, in subagents as well as here" — but the main session is using that tool fine. Use when: (1) you edited a file in `.claude/agents/` and dispatched that agent in the SAME session to confirm the fix; (2) a subagent returns "No such tool available: X" for a tool listed in its own `tools:` frontmatter; (3) a newly written agent file returns "Agent type '<name>' not found"; (4) you are about to escalate a subagent capability gap as a harness limitation or a settings problem. The agent registry — file list AND file contents — is resolved once at session start, so a correct repair reports as a failure and the failure names the tool rather than the staleness. Prevents a stale snapshot being escalated as a missing capability.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

container-green-host-red-detached-child-holds-tempdir

A hook or script self-test passes in a Linux CI/factory container and fails on the Windows host where the hook actually runs. Use when: (1) a Python self-test aborts with `PermissionError: [WinError 32] The process cannot access the file because it is being used by another process` at `TemporaryDirectory` cleanup, after the early checks passed; (2) `ignore_cleanup_errors=True` on the obvious tempdir does NOT clear it; (3) the code under test launches a detached child (`subprocess.Popen(..., start_new_session=True)`) and a test runs it with `cwd=<tempdir>`; (4) a Windows-only `endswith(".claude/state/x")` path assertion fails on backslashes. The child inherited the tempdir as its cwd and holds it for its lifetime; the fix is in the product (pin the child's cwd), not the test.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

private-steering-head-over-public-repos

Running one private repository that decides, audits and stages work for two or more public repositories it owns. Use when: (1) a question arrives that a sibling repo already answers and you are about to answer it from the steering repo instead; (2) you are about to build an instrument — script, metric, harness, gate, receipt schema, checker — without reading the siblings first; (3) you write or match a bare `#N` issue reference that will be read in a different repository than it was written in; (4) you are summing or averaging board health, drift or coverage across repos; (5) an artifact is produced and its home repo is unobvious; (6) a wayfinder-style map or index spans repos. Covers the cross-repo issue-number collision, per-board failure-mode asymmetry, the sibling-first prior-art gate, and artifact routing.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

squash-merge-absorbs-unpushed-base-commits

A squash merge silently relabels commits that were never yours to squash, when the branch was cut from a local base that sat ahead of its remote. Use when: (1) a PR's commit list shows commits you did not write on that branch — a release commit, a session close, a colleague's merge — alongside your own; (2) `git log` on the default branch no longer shows a commit you know landed, though its content is present; (3) a tag, a changelog entry, a state file or a handoff references a SHA that is no longer reachable from the default branch; (4) you are about to cut a feature branch and `git status` says "Your branch is ahead of 'origin/main' by N commits"; (5) a workflow commits to the default branch locally without pushing (a close ritual, a version bump, a generated-artifact commit) and later work branches off it. The content always survives; the absorbed commits' messages, SHAs and authorship do not.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

bash-cwd-drift-false-clean-grep

Catch the false all-clear produced when the Bash tool's persistent working directory drifts and a verification grep silently searches the wrong tree. Use when: (1) a `grep -rn` over known-present strings returns nothing and you are about to conclude the repo is clean, (2) an earlier `cd subdir/` in the session was followed by a repo-root-relative command, (3) a "no matches" result is being used as evidence rather than as an absence of evidence, (4) a stale-claim sweep or a pre-commit consistency check returns empty, (5) `ls`/`cat` on a root file reports No such file or directory. The Bash tool persists cwd across calls, so a grep with relative paths reports zero matches instead of erroring, and zero matches reads exactly like a passing check. Includes the guard that turns a silent miss into a loud one.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

fix-brief-consolidation-id-hygiene

Use when synthesizing findings from N parallel reviewers/seats/SMEs into a single fix-brief, disposition doc, or punch list. Prevents the "reviewer-local M-ID collision drops a finding" failure mode where each reviewer's local M1/M2/M3 numbering collides at the consolidation layer and one reviewer's finding gets silently dropped while another reviewer's same-numbered finding is kept under the consolidated label. Specifically: (1) N-seat parallel council fires producing per-seat findings, (2) multi-reviewer ai-slop / code-review / security-audit dispatches, (3) any synthesis step that flattens per-reviewer-local IDs into a global brief.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

github-linkcheck-404-throttle-false-negative

Diagnose CI link-checker (lychee and similar) failures reporting 404 on github.com issue/PR URLs that actually exist. Use when: (1) a link-check job goes red on `https://github.com/<owner>/<repo>/issues/<n>` links, (2) `gh api repos/<owner>/<repo>/issues/<n>` returns the issue fine, (3) the same check passed recently with no doc changes, (4) other unrelated CI jobs fail at the same time. GitHub throttles anonymous HTML page requests with 404 (not 429), which a checker cannot distinguish from a dead link; platform outages produce the same signature repo-wide.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

interactive-script-phantom-answers

A prompting script run without a TTY records fabricated answers instead of failing. Use when: (1) an interactive bash/python script "completed" but wrote empty or default values, (2) a wizard's summary says "wrote N values" yet the output file is blank, (3) `read -r` / `input()` returns instantly with no visible typing, (4) a y/N `confirm` resolved to its default that nobody chose, (5) you are about to tell a user to run an interactive script through an AI harness, a pipe, CI, or an editor "run" button. Covers the guard pattern for any script whose output becomes evidence.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

mutation-equivalent-in-architecture

Use when a mutation-testing campaign produces a surviving mutant that LOOKS like a genuine test-coverage gap but cannot be killed by ANY falsifiable test in the current code architecture. The honest disposition is RE-CLASSIFY as EQUIVALENT-IN-CURRENT-ARCHITECTURE and document the architectural constraint blocking independent killability — NOT ship a disjunction-based test that passes either way under the mutation (which is itself slop by the mutation-testing rubric's own discipline). Common trigger: sequential guards where guard N+1 can only fire when guard N also fires; mutating guard N+1's threshold is structurally unreachable so the surviving mutant is equivalent not gap. Sibling discipline to ai-slop-sentinel's "non-falsifying test" taxonomy: a test that passes under both current code AND the proposed mutation is exactly the slop pattern the mutation-testing rubric exists to surface — so writing one to "kill" the mutant defeats the entire purpose.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

openrouter-assistant-prefill-host-rejection

Diagnose and fix "This model does not support assistant message prefill. The conversation must end with a user message." errors when running Anthropic models through OpenRouter (opencode, agent harnesses, any client that sometimes ends a request on an assistant turn). Use when: (1) that exact error kills a run, branded [Azure], [Amazon Bedrock], [Google], or [Anthropic]; (2) the failure is intermittent across otherwise-identical runs; (3) a provider routing pin "fixed" it and it came back. Covers why pins cannot fix it, why intermittent green runs prove nothing, the behavioral probe that actually tests a candidate model, and the model-family change that resolves it.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

applied-layer-answer-hides-the-governing-result

Fires after a prior-art sweep SUCCEEDS: its sources are applied docs, not the theory they instantiate; you are about to record 'no prior art'; or a name-grep of zero is becoming an absence claim.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

structure-at-the-write-site

Fix for a downstream reader re-deriving a structural fact by pattern-matching prose that the program itself emitted. Use when: (1) a classifier greps an error/cause/log message for keywords to decide severity, retryability, or category; (2) a regex over a message string decides control flow, and adding a new message silently changes the outcome; (3) a status is inferred from an adjacent field that only correlates with it (a stage reached, a timestamp present, a count non-zero); (4) two consumers of the same message disagree about what it means. Root cause is that the site which KNEW the structural facts recorded only their prose summary, so every later reader must guess. Fix: record the facts as a value at the site that knows them, and delete the pattern match.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

agent-snapshot

Use when a subagent reports a tool as unavailable or an agent type as not found after you edited .claude/agents/ this same session. The registry is snapshotted at session start, so the fix is inert.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

blocked-form

Use when a guard, hook or sandbox refuses a destructive command and you are about to record the action as blocked, escalate it, or leave a ticket waiting. Read the refusal text first.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

brief-ids

Use when flattening findings from several reviewers or seats into one brief or punch list. Their reviewer-local M1/M2/M3 ids collide at the consolidation layer and one finding is dropped silently.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

cwd-drift

Use when a verification grep returns no matches and you are about to call the tree clean. The Bash tool's cwd persists across calls, so a relative path searches the wrong tree and reports zero.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

detached-child

Use when a self-test passes in a Linux container and fails on Windows with WinError 32 at TemporaryDirectory cleanup. A detached child inherited the tempdir as its cwd; fix the product, not the test.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

equivalent-mutant

Use when a surviving mutant looks like a coverage gap but no falsifiable test can kill it in the current architecture. Re-classify it as equivalent; a test passing either way is the slop it hunts.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

linkcheck-throttle

Use when a CI link checker reports 404 on github.com issue or PR URLs that gh api resolves fine. GitHub throttles anonymous page requests with 404, not 429, which reads exactly like a dead link.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

nameable-half

Use when writing or trusting a checkpoint band, handoff or packet that compresses a receipt. A summary reading "all X are P" where the source said "P or Q" keeps the half with the familiar name.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

phantom-answers

Use when a prompting script completes with nobody typing, writes blanks or defaults, or is about to run through a harness, pipe or CI. Without a TTY, read returns EOF rather than an error.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

prefill-rejection

Use when an OpenRouter run dies on "does not support assistant message prefill", intermittently, or after a provider pin appeared to fix it. Pins cannot fix it and a green run proves nothing.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

reviewer-horizon

Use when assembling a brief for a reviewer or subagent, or when its answer collides with a rule you never sent. Your selection became its design boundary, so the collision is manufactured, not found.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

skill-reachability

Use when a skill that is installed and working reports as missing, is refused by the Skill tool, or is absent from the listing. Three separate mechanisms produce that symptom; each has its own check.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

squash-absorbs

Use before cutting a branch while git says you are ahead of origin, or when a PR lists commits you did not write and a known SHA has left the default branch. Squash absorbs the unpushed base.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

steering-head

Use when one private repo decides and audits for public repos it owns: before building an instrument, when writing a bare #N issue reference, or when an artifact's home repo is unclear.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

uniform-eol

Use when a small edit produces a diffstat near the file's whole line count, or before writing a file with write_text on Windows. A uniform EOL conversion is invisible to a mixed-endings guard.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

write-site

Use when a reader greps an error, log or status message to decide severity, retryability or control flow. The site that knew the facts recorded only prose. Record them as a value, delete the match.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

wrong-altitude

Fires after a prior-art sweep SUCCEEDS: its sources are applied docs, not the theory they instantiate; you are about to record 'no prior art'; or a name-grep of zero is becoming an absence claim.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

anthropic-sdk-via-openrouter

Route Anthropic SDK calls through OpenRouter as a fallback when ANTHROPIC_API_KEY is absent but OPENROUTER_API_KEY is present. Pattern: anthropic.Anthropic(api_key=OPENROUTER_API_KEY, base_url="https://openrouter.ai/api/v1") + provider-prefixed model id "anthropic/claude-sonnet-4.6" (note dots, not dashes). The Anthropic SDK speaks Anthropic Messages API at the base_url; OpenRouter implements that contract at /v1/messages, so the same .messages.create(model=..., tools=..., tool_choice=...) call works without any code change beyond the constructor. Use when: (1) supporting users on Claude Code subscription auth (no direct ANTHROPIC_API_KEY but OPENROUTER_API_KEY available), (2) building any Python tool that uses the Anthropic SDK and wants to support OpenRouter-only environments without rewriting to use the OpenAI SDK or a different API contract, (3) implementing the symmetric fallback to a similar pattern on the OpenAI SDK side.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

exit-worktree-cwd-override-merge-from-worktree

When ExitWorktree refuses with "cannot be called from a subagent with a cwd override," merge the worktree branch back to main using `git -C "<main-repo-path>" merge <worktree-branch>` from inside the worktree, instead of trying to exit first. Git merge operates on branch refs and doesn't care about working directory. Use when: (1) you entered a worktree via EnterWorktree, did your work, and now want to merge back to main but ExitWorktree refuses with the cwd-override error; (2) you need to merge a branch in worktree A while you are physically cwd'd in worktree B; (3) any situation where the harness-managed cwd state blocks the directory change ExitWorktree would normally do. Avoids the trap of trying to chase the ExitWorktree refusal by navigating manually — the merge doesn't need the navigation.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

two-phase-doc-honesty-then-engineering

Response pattern after a state-drift catch surfaces BOTH doc inaccuracies AND a deeper engineering gap the docs were obscuring. Split the response into two phases that ship on different cadences: Phase A is an inline doc-honesty commit (truthful documentation of current state, orchestrator- direct, no authorization needed). Phase B surfaces the engineering gap to the PM as a separate dispatch decision (needs greenlight, costs more agent budget). Both ship; neither blocks the other. Use when: (1) a stale planning doc / spec / handoff catches mid-task and you're about to either silently re-do already-shipped work OR ship a doc-only fix that papers over a real engineering gap, (2) a "hard right over easy wrong" prompt forces you out of literal-spec adherence, (3) any catch where the documentation truth is small/fast and the engineering follow-on is larger/needs-authorization. Sibling to halt-as-deliverable (which covers the reframing of the work product); this skill covers the execution response after the catch

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

walk-the-recipe-as-target-user

Validation discipline for any "documented workflow" (README reproduction recipe, getting-started guide, install script, onboarding doc, deploy procedure). Simulate the ACTUAL target user's environment — different auth path, different SDK credentials, different OS, different tooling — and walk every documented step as that user would. Each step that breaks for the target user is real friction. Each step that silently assumes "you already have X set up" is an undocumented precondition. Use when: (1) the workflow was written by a developer whose machine has every privilege/credential/tool pre-configured and you're worried real users don't, (2) "first-touch hardening" or "on-ramp polish" or any phase whose stated purpose is making the workflow reachable for non-developers, (3) you've just changed the underlying CLI/API surface and the documented workflow predates the change, (4) a literal-spec adherence move (do the checklist items the doc names) would miss the friction the doc was MEANT to address. Catches the s

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

im-up

Use when a fresh Claude Code session must resume from an im-down packet and verify repository state before any work begins.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

closure-mode-at-boundaries

Closure-mode discipline at workflow boundaries. Use when a sprint, phase, or vertical slice locks clean, or when a multi-candidate "what next" decision arises. Owns the closure→build transition.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

skill-family-curation

Weeds a growing skill library: tests whether overlapping skills are one family using fixtures, then hands the human a disposition. Consolidation deletes child cards, so the human decides.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

im-down

Use when you are ending a Claude Code work session and the next session must resume from a verified packet instead of from conversational memory.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

honesty-first

Use when a drift catch surfaces both an inaccurate doc and the engineering gap it hid. Ship the doc truth inline now; surface the engineering work as its own dispatch decision. Neither blocks.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

sdk-via-openrouter

Use when ANTHROPIC_API_KEY is absent and OPENROUTER_API_KEY is present. Point the Anthropic SDK at openrouter.ai/api/v1 with a provider-prefixed model id; no other code changes.

0 Updated 2 days ago
MrBinnacle
DevOps & Infrastructure Listed

walk-the-recipe

Use when validating a README recipe, install script, onboarding or deploy doc written by someone whose machine already had everything. Walk every step in the target user's environment, not your own.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

worktree-cwd

Use when ExitWorktree refuses with a cwd-override error and the branch still needs merging. Merge by ref with git -C from inside the worktree; a merge needs no navigation.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

closure-mode

Closure-mode discipline at workflow boundaries. Use when a sprint, phase, or vertical slice locks clean, or when a multi-candidate "what next" decision arises. Owns the closure→build transition.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

click-clirunner-env-none-deletes

Click's CliRunner.invoke `env=` only overrides keys the dict names; an absent key is not deleted. Pass `{key: None}` to delete one. Use when a CLI test asserts on an env var being absent.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

halt-as-deliverable

When a pre-registration, pre-flight or sanity gate refuses to produce your deliverable, the catch is often worth more than the deliverable. Use when a discipline you built caught your own work.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

pretooluse-bash-guard-prose-false-positive

A PreToolUse Bash guard reads the whole command string, so it blocks prose that only mentions what it forbids. Use when a hook blocks its own install, or when writing a Bash matcher.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

router-skill-predicate-gap

A router rule can be live, healthy and match nothing anyone types: the pattern list omits the ordinary word, or a JSON escape left the pattern inert. Use before claiming a discipline is hook-enforced.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

success-test-accepts-any-output

A success check accepting any non-empty output passes on failure, because failure output is non-empty too. Use when a script reports OK but nothing happened, or a probe says NOT-FOUND batch-wide.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

git-pull-rebase-trap

Use before `git pull` where `pull.rebase` may be `true`. `--no-ff` does NOT stop a rebase under `pull.rebase=true` (silently ignored); it rewrites every local commit SHA. Check config first.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

github-pages-deploy-verification

Verify a GitHub Pages (or any CDN-fronted) deploy actually serves new content. Poll on a marker that did not exist pre-deploy; avoid sleep-and-curl chains the agent harness blocks.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

downstream-instruction-framing

Framing discipline for artifacts instructing a downstream reader — handoffs, plans, ADRs, subagent prompts. Separates evidence access from decision rights; marks which decisions evidence can reopen.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

concurrent-subagents-share-one-checkout-and-contend-on-head

Two or more subagents dispatched at the same repository by absolute path run in ONE working directory and contend on HEAD, so one agent's commit lands on another's branch and a rebase retargets a branch its author does not own. Use when: (1) about to dispatch more than one agent that will run git in the same repo, whether or not their FILES overlap; (2) a pre-launch check concluded "no collision" on the basis of file-overlap or different-repos reasoning; (3) `git log` on your branch shows a commit you did not author, or `git status -sb` reports ahead/behind counts that make no sense for a branch you just pushed; (4) an agent reports it rebased or committed and the result appears on someone else's ref; (5) `git branch -a --contains <sha>` returns only a branch belonging to different work. File-overlap analysis cannot see this class: the contention is on HEAD and the index, not on paths. The fix is worktree isolation at dispatch, not coordination after the fact.

0 Updated 2 days ago
MrBinnacle
Testing & QA Listed

mock-masked-stub-trap

A test that patches a helper which is itself a production stub reports green on a branch that never ran. Use when reviewing all-green tests on spend guards, idempotency, auth or refusal paths.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

subagent-research-reliability

Pre-dispatch, name the return channel — a subagent handback can be empty. Verify tool grants match the task. Post-return, verify every claim, negatives first.

0 Updated 2 days ago
MrBinnacle
Web & Frontend Listed

anti-slop-frontend-secure

Use when writing a single-file HTML artifact or a deployable frontend that must clear security gates - safe DOM construction, a host allowlist, a content-security policy, no secret in the output.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

parallel-review-disposition-schema

Use when dispatching parallel agents to verify/adjudicate findings — adversarial verify, security disposition, design triage, council fires. Shared disposition schema so seat outputs join cleanly.

0 Updated 2 days ago
MrBinnacle
Code & Development Listed

self-documenting-code

Use when reviewing code for clarity, refactoring names or flow without behavior changes, exposing units or side effects, or deciding which comments to keep.

0 Updated 2 days ago
MrBinnacle
Web & Frontend Listed

frontend-slop

Use when writing a single-file HTML artifact or a deployable frontend that must clear security gates - safe DOM construction, a host allowlist, a content-security policy, no secret in the output.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

shared-checkout

Use before dispatching more than one agent that runs git in the same repo, or when a commit or rebase lands on a branch nobody targeted. They share one HEAD, which a file-overlap check cannot see.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

clirunner-env

Click's CliRunner.invoke `env=` only overrides keys the dict names; an absent key is not deleted. Pass `{key: None}` to delete one. Use when a CLI test asserts on an env var being absent.

0 Updated 2 days ago
MrBinnacle
Testing & QA Listed

mocked-stub

A test that patches a helper which is itself a production stub reports green on a branch that never ran. Use when reviewing all-green tests on spend guards, idempotency, auth or refusal paths.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

pretooluse-prose

A PreToolUse Bash guard reads the whole command string, so it blocks prose that only mentions what it forbids. Use when a hook blocks its own install, or when writing a Bash matcher.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

pull-rebase

Use before `git pull` where `pull.rebase` may be `true`. `--no-ff` does NOT stop a rebase under `pull.rebase=true` (silently ignored); it rewrites every local commit SHA. Check config first.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

stale-deploy

Verify a GitHub Pages (or any CDN-fronted) deploy actually serves new content. Poll on a marker that did not exist pre-deploy; avoid sleep-and-curl chains the agent harness blocks.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

vacuous-check

A success check accepting any non-empty output passes on failure, because failure output is non-empty too. Use when a script reports OK but nothing happened, or a probe says NOT-FOUND batch-wide.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

dead-predicate

A router rule can be live, healthy and match nothing anyone types: the pattern list omits the ordinary word, or a JSON escape left the pattern inert. Use before claiming a discipline is hook-enforced.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

decision-rights

Framing discipline for artifacts instructing a downstream reader — handoffs, plans, ADRs, subagent prompts. Separates evidence access from decision rights; marks which decisions evidence can reopen.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

disposition-schema

Use when dispatching parallel agents to verify/adjudicate findings — adversarial verify, security disposition, design triage, council fires. Shared disposition schema so seat outputs join cleanly.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

subagent-handback

Pre-dispatch, name the return channel — a subagent handback can be empty. Verify tool grants match the task. Post-return, verify every claim, negatives first.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

rowless-card

Fixture card whose EVIDENCE.md is missing one required row.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

poison-card

Fixture card missing one required file.

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

skill-necessity-gate

fixture

0 Updated 2 days ago
MrBinnacle
AI & Automation Listed

meta-fixture-card

fixture

0 Updated 2 days ago
MrBinnacle

Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.