BryceEWatson
UserA curated collection of Claude Code skills — code-review loop, local chat-history search, session end/resume, Gemini image generation, transcript retrospectives — that run machine-wide with just python and node.
Categories
Indexed Skills (16)
fable-seat
Operating profile for a Claude Code session whose model is Fable (`claude-fable-5`) — the dials that make a Fable-led session run cheaply. Effort defaults to `high` and steps DOWN for routine work (not `xhigh` by reflex); task framing leads with the outcome instead of prior-model step-by-step scaffolding; the prompt cache stays warm by farming cheap work to a subagent rather than inline-switching the main loop's model; long runs get a task budget plus context compaction; and durable state lives on the memory surface. Apply ONLY when the session model is Fable — it is a genuine NO-OP on Opus and every non-Fable model. This is Fable-mechanics guidance, NOT a superiority claim (it does not assert Fable beats Opus). Use it when you're running Fable and want to set the operating dials, before kicking off a long agentic run on Fable, or when asked "how should I run Fable", "Fable operating profile", "Fable seat", "calibrate for Fable", "am I running Fable cost-efficiently".
gemini-image
Generate and edit images with Google's Gemini models (Nano Banana, Gemini 3 Pro Image, 3.1 Flash Image) from a zero-dependency Python CLI. Use to create, edit, restyle, or compose an image with Gemini or Google AI, including image-to-image with reference images. Also handles text and model-listing. The maintained successor to the older `gemini-client` plugin.
global-review-loop
Mine ALL of your local Claude history across EVERY project (both corpora — Claude Code CLI and Cowork local-agent-mode) for operator friction that recurs across MULTIPLE projects, then propose GLOBAL ~/.claude leverage (a CLAUDE.md rule, a new/edited global skill, a memory, or a settings hook) to fix it — each proposal reconciled against what already ships globally, self-validated by an adversarial /review-loop --mode claim loop, and recorded in a cross-run ledger so nothing is re-proposed, dismissed-twice, or shipped on its own say-so. DISTINCT FROM: signal-scan (one repo → that repo's .claude/ config), command-retro (Command's own logs → Command's queue), and transcript-analysis (ONE project, Cowork-only, proposes that project's CLAUDE.md candidates). This skill is the GLOBAL, cross-project layer: it only surfaces friction proven to recur across >=N DISTINCT projects (counted by cwd-PATH attribution only) and only ever proposes changes to ~/.claude. Invoke as /global-review-loop. Triggers: "what global rule
monitor-agent-thread
Monitor a live or recent Claude Code or Codex thread from the other product using local append-only session logs, safe event extraction, Git/worktree evidence, and recurring heartbeats. Use when the user asks to watch, babysit, follow, monitor, check progress on, or notify them about another Claude Code/Codex session, including Claude-to-Codex and Codex-to-Claude monitoring.
pattern-retrospective
Apply rigorous pattern-retrospective analysis to Claude transcripts. Fires on any study, blog post, handoff doc, retrospective, spec, or improvement proposal that draws conclusions from past Claude sessions. Enforces target-system audit BEFORE specifying its requirements, streaming JSONL parsing, 5-tuple extraction with provenance tags, self-falsification of every cited claim, Bayesian confidence smoothing, and correlation-not-causation discipline. Prevents the credulous-handoff failure mode where ~60% of proposed requirements already exist in the target system.
review-loop
Dispatches a multi-agent review team over this session's work, runs an execution-grounded lint/test/build check, falsifies each finding against the diff, fixes load-bearing issues, and re-reviews until clean or the budget hits. Always records a commit-pinned verdict on the branch's open PR, so the review leaves a trail. Invoked manually or by the Stop hook after real work outside plan mode. `--mode claim` reviews an analytical conclusion against its primary sources instead of a diff; `--mode deliverable` reviews finished reader-facing work (a report, a content page, a decision briefing) for whether the human it is handed to can actually use it.
session-pickup
Start a continued Claude Code session by rehydrating from a prior session's handoff, the inverse of session-end. Reads the latest handoff in .claude/handoffs/ and what it points to, reconciles it against current git and file state to catch drift, rebuilds the todo list, settled constraints, and open verification debts, then presents the immediate next action and confirms before acting.
session-portal
Safe bidirectional message portal between Claude Code and Codex sessions. A durable local SQLite queue with an authenticated stdio MCP interface lets one session drop a short, authored message into another session's inbox; delivery is pull-only, so the recipient itself picks it up at a safe turn boundary, acknowledges, and acts under its own permissions. Identity is derived from a bearer token (not caller-asserted) and elevated actions need operator grants. Never writes transcripts, never pushes into or resumes another session, never runs message content. Use to hand a follow-up from Claude Code to Codex (or back) without manual relay. Monitoring stays separate and read-only (monitor-agent-thread).
transcript-analysis
Analyze Cowork session transcripts for recurring narrative patterns — repeated instructions, planning loops, failure patterns, decision re-litigation, and knowledge re-establishment. Produces a structured report with CLAUDE.md candidates, unresolved loops, and pattern frequency data. Use this skill when the user wants to review what keeps repeating across sessions, extract lessons from past work, audit their collaboration patterns, or identify rules that should be encoded permanently. Triggers on: transcript analysis, session analysis, narrative analysis, what keeps repeating, review our sessions, cowork analysis, pattern analysis, what should go in CLAUDE.md, extract lessons, session retrospective, retro, session patterns.
weekly-work-log
Build and refresh the brycewatson.com "Weekly Work Log" report page (/weekly-report, archived under /log) for the last completed week. Discovers the week's work from session-end handoffs + git, distils it into honest public-voice items, re-derives and verifies every number, and (unattended, every Sunday night via a Claude scheduled task) opens a judgment-assisted review PR. Never publishes unattended.
ship-it
Take a build task from first read to a shipped, reviewed PR — the full research → plan → review-the-plan → implement → test → PR → review-the-PR chain, looping each checkpoint until the definition-of-done holds. Fires on any end-to-end build intent: an issue to take all the way, or a freeform goal stated with the whole chain. Unlike typing the chain by hand, it adversarially reviews the PLAN before code is written, gates plan-approval through the operator, exercises the REAL app, and never silently de-scopes.
honestweek
Turn a completed week of your AI coding sessions into an honest, git-verified, private-by-default work summary. Run /honestweek to scaffold config, discover the week's sessions into a redacted digest, distil it into reviewable work items with a status badge and receipt, build it with verify-or-abort, and review the draft before you publish it yourself. Automatic session-derived digest items carry receipts without claiming work status. honestweek never auto-publishes.
chat-arch-thrash-detect
Calibration-gated PostToolUse hook that watches the rolling window of tool calls in a Claude Code session and emits a stderr nudge when it detects an Edit-thrash, Read-loop, Test-loop, or Tool-flail pattern. NOT a user-invokable skill — the directory exists to host the hook script + install/uninstall machinery alongside the rest of the chat-arch skill family. Gated behind `CHATARCH_THRASH_DETECT=1` for the 4-week calibration window (plan §Phase 4 #8).
chat-history-search
Exhaustive search across ALL of Bryce's local Claude chat history, both Cowork and the Claude Code CLI, to find past prompts, recover a conversation, inventory how often a pattern was used, or audit prior work across projects. Knows every log location, every JSONL line shape, and the false-positive gotchas that trip up naive grep.
seo-index-validation
Validate whether a DEPLOYED site is being crawled and indexed correctly, and diagnose plus fix why pages are not indexed. Use on a Google Search Console alert about non-indexed pages, when checking index health after a deploy or migration, or to confirm a prior indexing fix cleared. Probes live HTTP, reads the GSC report via the browser, fixes at the correct layer under confirmation-gating, then schedules a lagged re-check. NOT for keyword research, content optimization, or backlinks.
session-handoff
Alias of `session-end` (this skill was renamed). Ending / closing out a session — and handing it off when work is mid-flight — now lives in session-end; "handoff" is its mid-flight mode. Invoking this just routes to session-end so old muscle memory ("/session-handoff") keeps working.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.