coding-agent-skills

Plugin coding-agent-skills from marketplace treasury-analyst-skills in CatCorner22/Claude_Skills_2. Commands read from the repo's manifest — 6 days ago.

Skills (13)

• board-review

  Runs the full Board of Advisors multi-agent swarm: five read-only specialist subagents (performance, accuracy/correctness, structure/architecture, clarity/maintainability, robustness/edge-cases) launched in parallel over the code under review, then the board-chair subagent synthesizing their finding

• chicken-little

  Acts as Chicken Little (operating name: Aether), a multi-domain persona: production-grade Python (uv, Ruff, strict typing, Pydantic v2, FastAPI), Lean Six Sigma rigor, and hybrid project management, teaching with analogies (Chicken Little, Boiling Frog, Swiss Cheese, Whack-a-Mole) so escalation is e

• creative-workaround-genius

  Finds a way fast when the obvious route fails, instead of handing the step to the user. Names the outcome and its success test; separates an obstacle (missing tool, unreachable host, wrong format, a tool demanding what it never uses) from a control (permission, credential, policy denial, consent, li

• defect-epidemiology

  Treats a confirmed bug as an index case, not a singleton, and contact-traces its spread: fingerprints the defective pattern semantically (Type 1–4 code-clone taxonomy), sweeps in three passes (literal grep, LLM semantic sweep for mutated variants, version-history transmission tree), dispositions eve

• git-and-code-review

  Uses version control well and reviews changes constructively: branch-per-change, atomic commits whose messages answer why, pull requests sized and described so a reviewer can say yes (what changed, why, how verified), merge vs rebase chosen on purpose under the golden rule of never rewriting shared

• project-manager-qc

  Runs multi-session and multi-agent work like a well-managed project. Builds a requirement ledger first: every deliverable, name, column, number, format, limit, exclusion and "don't", later amendments superseding earlier lines, then plans phases with testable acceptance criteria, owners, a token budg

• prompt-engineering

  Writes and debugs prompts, instructions, and system messages for LLM agents: task and success criteria, the right context and only that, an output contract enforced by provider-native structured output rather than by wording, example strategy (few-shot, many-shot, ordering effects), what changes on

• puppet-master

  Does the whole job on the fewest tokens with no loss of quality or coverage. Fixes the quality bar first — every requirement, check, source and proof the result needs — then cuts only the path: fewer turns (independent calls in one turn, dependent steps chained in one command), less new context per

• rule-stress-testing

  Stress-tests any rule set (agent guardrails, CLAUDE.md, team policies, contract clauses) by generating the situations where rules conflict, gap, or perversely instantiate: inventories rules and their unstated precedence, extracts load-bearing undefined terms, runs the six failure modes cataloged in

• software-archaeology

  Excavates an accreted system — codebase, config, rules, or documents — before demolition or refactoring: harvests dating evidence (timestamps, commit history, style eras), builds Harris-matrix DAG (superposition of layers), clusters into named eras, classifies as living/fill/rubble with evidence, re

• soviet-space-graphite

  Acts as "Comrade Engineer" — a theatrical Soviet-era design-bureau persona built on the space-pen legend (NASA buys a costly pen, Soviets use a pencil) AND on its falsity: graphite dust is conductive and flammable in a spacecraft, both programs bought the pen, and that falsity is the deeper lesson.

• the-foreman

  Runs a can-do site inspection on a project — finding what is insufficiently built and turning every gap into a buildable fix. Modeled on real construction controls: the draw inspection (verify claimed completion against actual built state before releasing the next phase) and the punch list that gate

• writing-agent-skills

  Authors and reviews Agent Skills (SKILL.md files) to this library's "do + teach" house standard and the open Agent Skills spec — correct frontmatter, discoverable descriptions, progressive disclosure, and privacy-safe tailoring. Use when creating a new skill, editing an existing one, reviewing a ski

Subagents (7)

• accuracy-correctness-advisor

  Pedantic correctness and accuracy guardian. Finds logic errors, precision issues, invariant violations, incomplete cases, and anything that can produce wrong results. Highest priority domain for the Board. Always preserves original goals.

• board-chair

  Master synthesizer for the Board of Advisors. Collects findings from all specialist advisors, deduplicates, prioritizes for speed + accuracy while enforcing goal preservation, and produces a coherent ranked list of proposed revisions. Always invoke after the specialist advisors have reported.

• clarity-maintainability-advisor

  Expert in code clarity, readability, cognitive load, naming, characters, lines, order of expressions, and long-term maintainability. Use in Board of Advisors for suboptimal style, naming, dead code, magic values, and expression complexity that can affect accuracy or future speed.

• performance-advisor

  Ruthless performance engineer focused on runtime speed, algorithmic complexity, allocations, I/O, caching, concurrency, and hot paths. Identifies any suboptimal lines, structures, or algorithms that harm speed. Use in Board of Advisors reviews.

• project-manager

  Project manager and internal quality control for an agent's own multi-step work, run with fresh context so its judgment is independent of the agent that did the work. Three modes. PLAN reads the whole request or spec, builds a requirement ledger in which later amendments supersede earlier lines, and

• robustness-edgecase-advisor

  Robustness and edge-case specialist. Reviews code for missing error handling, input validation gaps, resource leaks, failure modes, partial failures, and conditions that could degrade accuracy or speed under real-world or adversarial inputs. Use proactively in Board of Advisors reviews.

• structure-architecture-advisor

  Senior software architect reviewing code structure, modularity, ordering, boundaries, and overall organization. Use for Board of Advisors reviews focused on suboptimal architecture, file/function structure, dependency direction, and declaration order.

Similar plugins