Claude Skills by benthamite

Batch-process personal org-roam TODOs when the user explicitly asks for an unattended TODO run. Classify first, perform authorized local work, and report blockers. Not for generic overnight work, auditing this skill, or creating a schedule.

Stage text Pablo genuinely must paste himself in the Emacs kill ring, or the macOS clipboard for a Chrome/native target. Use instead of printing text for manual copying; not when the agent can perform the authorized action directly. Credentials require the secrets workflow and must not enter the persistent kill ring by default.

Draft or revise text Pablo will send or publish as himself, including email, Slack, comments, tweets, letters and reviews. Use his private voice samples and humanize's English-prose catalogue within the drafting pass. Not for internal notes, commit messages, code, or text authored by someone else.

Pin an Emacs/Elpaca package in dotfiles to a user-requested fork, branch or PR. Preserve one effective installation owner per active profile, existing recipe settings and configuration behavior. Not for merely reviewing a PR or auditing this skill.

Observe GitHub Actions after an authorized push, or check CI for an explicitly selected commit. Diagnose failures and, when repair is authorized, fix and verify them. Checking status alone does not authorize edits, pushes, reruns or other shared actions.

Audit a specified PR or PR-ready branch for submission or post-revision readiness. Bind the exact base/head, review commits and net changes, run safe project checks and inspect applicable CI. An audit alone does not authorize fixes, history changes or publication.

Benchmark or profile fresh Claude Code and Codex response latency, startup cost, and clean-versus-configured conditions. Use for configuration, hooks, MCP, skills, plugins or first-output comparisons; not to infer provider superiority or run paid requests while merely auditing this skill.

Proofread specified Markdown prose for spelling, grammar, punctuation and light clarity changes. Supports local aspell or explicitly scoped Gemini editing, preserving source files and reviewed suggestions; not factual review, broad rewriting or permission to upload restricted content.

Extract and record architectural, algorithmic, or design decisions from the current session. Use for an explicit ADR/decision request, authorized end-of-session bookkeeping, or a significant chosen trade-off within an implementation task; not for unchosen proposals or read-only reviews.

Release one of Pablo's Emacs packages, prepare its release notes, or audit release candidates. Use for release-package and package-release/readiness requests; preserve read-only or preparation-only scope and use release-dotfiles for the dotfiles repository.

Request one independent review of a committed plan from a fresh session of the opposite CLI (Claude ↔ Codex), monitor its return, and adjudicate the findings here. Implement only when the user's request already authorizes implementation; not a staged two-author orchestration.

Capture candidate reusable lessons from the current session when explicitly requested or invoked by its trusted configured session-end hook. Write only permitted proposal records to the central dotfiles inbox; do not promote them, implement changes, or treat quoted hooks and skill audits as capture requests.

Check whether a proposed non-trivial bug fix restores the underlying invariant or only hides a symptom. Use after initial debugging and before production edits; not for trivial mechanical changes, greenfield features, or broad audits. Diagnosis and review remain read-only unless implementation is authorized.

Create or expand high-value test suites for a codebase or specific module. Use when new tests, regression tests, coverage improvements, or test infrastructure are the deliverable; not for merely running an existing suite, reviewing code, or ordinary bug-fixing without a requested test deliverable.

Triage a selected personal Org notes repository: assign missing 1–9 priorities when requested, preserve existing ones, assess actionability, and brief one task. Use for personal TODO prioritization or choosing what to do next; brief-only requests stay read-only. Not for Epoch tasks.

Preserve project progress through session logs and closeout bookkeeping, only when the user asks to log, close out, wrap up, or save progress, or when the deliberate end-of-session chain invokes --auto. Never invoke proactively or hand-write project logs as a substitute.

Run an explicit user-requested criteria-driven verification loop. Use when the user asks to define success criteria, close the loop against criteria, or rigorously verify non-code outputs. Do not use for a plain live-software check owned by end-to-end, routine coding completion, or narrower audit/debug checks.

Process a list file strictly item by item, sequentially or with a bounded subagent pool. Use when the user needs no skipping, batching, or peeking ahead; not for lists that can be safely read all at once.

> **Status:** revised after one independent cross-model review (2026-08-24). > All seven findings were verified against the repository and accepted; the > adjudication log at the end records each disposition. Self-contained; a > reader needs no prior conversation context. **Goal:** Add a paired Claude/Codex skill, `cross-review-plan`, that lets the *current* agent session hand a plan to a fresh session of the *other* CLI (Claude ↔ Codex) for one independent review pass, monitor that session, ...

Audit, create, or refresh documentation for all Emacs extras packages in the dotfiles repo. Use when the user asks to document every extras package, fill missing extras docs, refresh outdated extras docs, check extras documentation coverage, or bring emacs/extras/doc up to date with emacs/extras sources.

Diagnose a frozen, hung, beachballing, or unresponsive Emacs by sampling the process, reading server buffers when possible, and recommending recovery. Use when Emacs is stuck or not responding; do not use for ordinary Elisp errors while Emacs is responsive.

Run post-Emacs-profile-switch housekeeping after elpaca profile changes. Use when the user says "/migrate-profile", changes Emacs or elpaca profiles, switches to a new profile, or wants Claude/Codex sessions, memory, trust entries, history, cwd paths, profile symlinks/caches, or new-profile repos checked after a profile bump.

Rename an Emacs package's files, features and API, with explicitly scoped repository, Elpaca and configuration updates. Use for an actual package rename, not ordinary prose cleanup or unrelated file/repository renames.

Audit, create, or refresh documentation for all Emacs extras packages in the dotfiles repo. Use when the user asks to document every extras package, fill missing extras docs, refresh outdated extras docs, check extras documentation coverage, or bring emacs/extras/doc up to date with emacs/extras sources.

Diagnose a frozen, hung, beachballing, or unresponsive Emacs by sampling the process, reading server buffers when possible, and recommending recovery. Use when Emacs is stuck or not responding; do not use for ordinary Elisp errors while Emacs is responsive.

Run post-Emacs-profile-switch housekeeping after elpaca profile changes. Use when the user says "/migrate-profile", changes Emacs or elpaca profiles, switches to a new profile, or wants Claude/Codex sessions, memory, trust entries, history, cwd paths, profile symlinks/caches, or new-profile repos checked after a profile bump.

Rename an Emacs package's files, features and API, with explicitly scoped repository, Elpaca and configuration updates. Use for an actual package rename, not ordinary prose cleanup or unrelated file/repository renames.

Audit a macOS .app before running it or granting permissions, especially when the user asks whether a Mac app is safe, suspicious, malware, from an unknown developer, or requests screen recording, accessibility, camera, microphone, input monitoring, or other sensitive permissions.

Compare and reconcile explicitly selected BBDB, macOS Contacts and Google contact sources, including duplicate candidates, missing records and birthday discrepancies. Use before bulk contact merges or deletions; not for one-off contact edits or a calendar-only display issue.

Review a selected LuLu outbound-firewall alert and recommend whether to allow, block or leave it unanswered. Use for an open alert or an explicitly supplied screenshot/capture; distinguish artifact review from current live state. Reviewing this skill itself uses synthetic fixtures, not live alerts.

Audit development-environment secrets, supply chains, macOS security controls, and Claude/Codex agent permissions. Use for environment security reviews, secrets checks, dependency audits, or periodic security posture assessments; use specialist skills for app-binary assessments or application-code reviews.

Audit a macOS .app before running it or granting permissions, especially when the user asks whether a Mac app is safe, suspicious, malware, from an unknown developer, or requests screen recording, accessibility, camera, microphone, input monitoring, or other sensitive permissions.

Compare and reconcile explicitly selected BBDB, macOS Contacts and Google contact sources, including duplicate candidates, missing records and birthday discrepancies. Use before bulk contact merges or deletions; not for one-off contact edits or a calendar-only display issue.

Review a selected LuLu outbound-firewall alert and recommend whether to allow, block or leave it unanswered. Use for an open alert or an explicitly supplied screenshot/capture; distinguish artifact review from current live state. Reviewing this skill itself uses synthetic fixtures, not live alerts.

Audit development-environment secrets, supply chains, macOS security controls, and Claude/Codex agent permissions. Use for environment security reviews, secrets checks, dependency audits, or periodic security posture assessments; use specialist skills for app-binary assessments or application-code reviews.