Skip to content

google/elephant-goldfish-model / skills/egm-lite

egm-lite

Verified

Executes a lightweight 5-step EGM (Elephant-Goldfish-Model) protocol for small, rapid iterations, specification revisions, code refactors, and bug fixes. Use when the user invokes /egm-lite, issues a phase command (checklist, exec, review, goldfish), or resolves a subagent review response. Don't use for large multi-component features requiring a formal design doc (use egm-brainstorm).

FileKindSize
SKILL.mdskill md23.3 KB
TEST.mdmarkdown4.8 KB
agents/openai.yamlconfig194 B
assets/icon.svgother477 B
test_scenarios.yamlconfig33.9 KB
# Test Plan: EGM-Lite Skill

Frontmatter, link checks, and scenario coverage run in `pytest tests/unit` (see [TESTING.md](../../TESTING.md)).

## Protocol Verification Checklist

When evaluating or exploring `egm-lite` interactively in chat, verify:

-   **Step 1 (Critique Pause):** The agent adopts the **Elephant (Critic)**
    persona, leads with a plain-English `> **BLUF:**` blockquote (followed by
    `---`) on whether the approach is ready for `checklist` or blocked,
    challenges assumptions without editing repository code, documents competing
    approaches in `egm_lite_alternatives.md` without inline discussion when
    multiple viable options exist (omitting it for trivial or single-path
    tasks), and prompts for `checklist` only, answering `exec` with a state advisory.
-   **Step 2 (Checklist & Spec Gating):**
    -   *Section 2A (Checklist Generation):* Upon `checklist`, the agent adopts
        the **Elephant (Planner)** persona, outputs an itemized checklist of
        change targets (with files and referenced specs as clickable file and
        line-range links), rationales, and any user-agreed pre-review exclusions
        as GFM task list checkboxes (`- [ ]`), evaluates deterministic
        complexity criteria (two or more production code files,
        async/concurrency, public APIs, or `[⚠️ UNVERIFIED]`/`[⏳ ACK]` tags),
        emits a BLUF that recommends the spec review (`goldfish` / `review`) on
        non-trivial tasks or says direct execution is reasonable on trivial/doc
        tasks, and in every case offers both `exec` and `goldfish` / `review`.
    -   *Section 2B (Stateless Spec Review):* Upon `goldfish` or `review`
        pre-implementation, the agent adopts the **Goldfish (Specification
        Reviewer)** persona via the harness's fresh-context subagent tool
        (halting if none is available), injects checklist markdown and the Spec
        Review Rubric (skipping pre-review exclusions and accepted `CONTEST`
        items, and forbidding AST/diff checks), and halts to await the response.
-   **Step 3 (Exec Protocol):** Upon `exec` (or clicking **Proceed**), the agent
    adopts the **Elephant (Builder)** persona. If following a verified
    specification, it reads `egm_lite_checklist.md` directly without re-creating
    it; if true fast-path, it initializes the checklist first. It implements
    only approved checklist items (covering code, specifications, or design
    docs), runs local tests covering affected targets (or N/A / linter for pure
    doc edits), marks completed items with `- [x]`, leads with a `> **BLUF:**`
    on escalations, self-repairs, or multi-file summaries (omitting on clean
    single-target runs), and gates on prompting for `review` or `goldfish`.
-   **Step 4 (Adversarial Diff Review):** Upon `review` or `goldfish`
    post-execution (or with no task in progress on an existing diff), the agent
    adopts the **Goldfish (Reviewer via Subagent)** persona (spawning via the
    harness's fresh-context subagent tool or stopping if none is available),
    restricts findings to material issues while suppressing stylistic nitpicks,
    enforces a strictly binary review response (findings only vs. explicit
    `[✅ LGTM]` on zero findings), evaluates against the Diff Review Rubric
    (adapted for documentation diffs by replacing runtime checks with factual
    consistency and markdown hygiene), and halts to await the subagent's
    response.
-   **Step 5 (Unified Post-Review Resolution Gate):** Upon receiving subagent
    findings (spec or diff review), the agent adopts the **Elephant (Resolver)**
    persona, keeping triage presentation-only without file mutations, opens with
    a plain-English `> **BLUF:**` blockquote (followed by `---`) led by
    **`[🛠️ X FIX | 🛡️ Y CONTEST]`** (or **`[✅ LGTM]`** on a clean pass), triages
    findings into continuous numbered and priority-tagged `[X] [PY]` items
    (`P0`..`P3`), branches on the review that responded (Pre-Implementation spec
    triage offers `checklist`, `exec`, or `goldfish` re-review and strictly
    forbids premature task closure; Post-Implementation diff review triage
    offers `checklist` or `exec`), and persists accepted `CONTEST` rationales in
    `egm_lite_checklist.md`.
-   **Top-Level Invariants & Graceful Recovery:** `goldfish` and `review` are
    interchangeable aliases bound to state, invalid state leaps (e.g. diff
    review with nothing to review, spec review before checklist creation) emit a
    graceful `> **BLUF:** [State Advisory]` blockquote, and fast-path execution
    natively supports iterative specification and design doc updates.
-   **Code References & Non-Sycophantic Posture:** Referenced local files and
    line ranges use clickable links, internal operational directives are never
    echoed into user chat, and responses lead with technical risks or defects
    without flattering adjectives or polite apologies.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall