code-goblin-pro

Plugin code-goblin-pro in rcmoynihan/codex-skills. Commands read from the repo's manifest — 1 month ago.

Skills (24)

• agent-orchestration

  Choose and coordinate agent architectures, from a single agent with tools up to full multi-agent orchestration. Covers when one agent suffices, the core orchestration patterns (sequential, concurrent, router, group chat, handoff, hierarchical, magentic), composable techniques (reflection, best-of-N,

• auto-resolve-pr-comments

  Autonomously resolve a PR's review comments end to end with no human review of the plan — runs plan-pr-comment-responses to produce the response plan, then immediately runs execute-pr-response-plan to implement, validate, verify, push, drive CI green, and post/resolve replies, then runs learn-from-p

• code-review

  Multi-persona code review of a branch or PR. Resolves the changes (current branch's PR, or a given PR / branch), spawns a review-lead subagent that runs always-on plus diff-warranted reviewer personas in parallel, and writes a single markdown review report to the plugin's temp dir. Report-only — nev

• coding-agent-pitfalls

  Common failure modes of AI coding agents (Claude Code, Codex, and similar) — doing more than asked, overengineering, solving the wrong problem, acting on shaky understanding, disrespecting the codebase, and mis-verifying. Use when planning or executing a coding task and you want to check your approa

• design-grill

  Grill the technical side of a compiled spec relentlessly — architecture, data/state, interface realization, execution, ops, deployment — one question at a time, recommending an answer for each, while keeping a durable on-disk agenda. Holds every element of the spec as locked input; resolves high-lev

• drive-claude-code

  Delegate coding tasks to the Claude Code CLI (Anthropic's autonomous coding agent) from the shell — one-shot print-mode runs or multi-turn interactive tmux sessions for building features, refactors, and PR reviews. Use when the user asks to drive, orchestrate, or delegate work to Claude Code as a su

• drive-codex

  Delegate coding tasks to the OpenAI Codex CLI (codex exec) from the shell for building features, refactoring, PR reviews, and batch issue fixing. Use when the user asks to drive, orchestrate, or delegate work to Codex as a sub-agent.

• execute-pr-response-plan

  Execute a response plan produced by plan-pr-comment-responses — implement every comment's TODOs, run the project's tests/validation until green, have Claude Code independently verify the work matches the plan with no scope creep, commit + push, drive PR CI to green, then post the drafted replies and

• handoff

  Compact the current conversation into a handoff document for another agent to pick up.

• hello-world

  A minimal smoke-test skill. Use when the user says "hello world" or asks to test/verify that the skills plugin is installed and working.

• interrupt-resume

  Resume a task that was interrupted before it finished. Use when the user says "resume", "continue", "pick up where you left off", or otherwise wants you to carry on from an interrupted task.

• learn-from-pr-comments

  Turn the issues a PR's reviewers raised into minimal, de-duplicated rule additions for the code-review reviewer profiles, so we catch the same class of issue ourselves before the next human review. Reads the response plan (and any resulting diff), keeps only comments that generalize, verifies existi

• opp-grill

  Interview a problem or opportunity that has no articulable solution idea yet — one question at a time, recommending an answer for each, holding every claim to a stakes-calibrated evidence bar — then generate candidate solution directions and end at a commit gate whose verdict is the user's: go / no-

• plan-pr-comment-responses

  Turn a PR's diff and reviewer comments into a concrete, comment-by-comment plan for addressing them — per-comment decision (implement / alternate / push back / defer), implementation TODOs, drafted PR replies, a verification plan, and a re-review checklist. Plans only; does not edit code or post rep

• post-review-comments

  Take a completed code-review markdown report (from the code-review skill) and post its findings to the PR as comments — strongly preferring inline comments anchored at each finding's file:line, with at most one top-level comment reserved for a genuine large cross-cutting concern. Never posts a summa

• prior-art-check

  Check whether an existing solution already satisfies a compiled spec before the design phase is grilled — the spec-keyed buy-vs-build gate. Reads spec.md, surveys candidate solutions (internal, open-source, commercial) via the Scout, scores each against the spec's invariants, load-bearing requiremen

• prompt-engineering

  Write strong prompts for LLMs — direct model calls, single tool-using agents, and multi-agent systems. Covers the core stance (specificity and completeness over brevity), prompt anatomy and ordering, the high-leverage techniques (be specific, give context and the why, positive instructions, multisho

• review-and-comment

  Autonomously review a branch/PR and post the findings as PR comments end to end, with no human review of the report in between — runs code-review to produce the review report, then immediately runs post-review-comments to post the findings inline (top-level only for a genuine cross-cutting concern,

• spec-grill

  Grill the product side of an idea relentlessly — one question at a time, recommending an answer for each — while keeping a durable on-disk agenda so the interview follows tangents without getting trapped by them. Exclusively non-technical — what it is, who it's for, what it must do; technical asides

• state-it-plainly

  Restate your own previous response in plain, jargon-free language — the bottom line, the key decisions, and what actually matters — so a long or noisy answer becomes easy to understand at a glance. Manual invocation only.

• survey-terrain

  Survey the existing product, codebase, and organization that future work lands in, and compile a durable terrain pack — a lightweight map/index/glossary (glossary, product surface, system map, conventions & paved roads, constraints & gates, internal prior art) written to <repo>/.grill/terrain.md. Th

• to-design

  Turn a completed design-grill session + its to-spec PRD into a single, complete technical design document in the grill run dir — the *how* that realizes the spec's *what*: architecture, data/state, interfaces, execution, cross-cutting concerns, deployment/rollout, technical decisions (ADRs), risks,

• to-idea

  Formalize a completed opp-grill session (or the current conversation) into a single idea brief markdown file in the grill run dir — the problem/opportunity frame with its evidence grades, the considered directions, and the commit verdict (go / no-go / defer / adopt). Deliberately half a spec; on a g

• to-spec

  Formalize a completed spec-grill session (or the current conversation) into a single product PRD/spec markdown file in the grill run dir — the *what*: product framing (problem, solution, scope) and the governing invariants, followed by a rigorous, testable statement of user stories, requirements, ex

Similar plugins