From silver-bullet
Runs 2–5 focused feasibility experiments with hypothesis, code, and verdict before committing to an implementation approach.
How this skill is triggered — by the user, by Claude, or both
Slash command
/silver-bullet:silver-spike <technical question or approach to investigate> [--quick] [--wrap-up]<technical question or approach to investigate> [--quick] [--wrap-up]The summary Claude sees in its skill listing — used to decide when to auto-load this skill
SB-owned spike workflow. Use when you need evidence before choosing an
SB-owned spike workflow. Use when you need evidence before choosing an implementation strategy — not research notes, but running code that proves or disproves an approach. Each experiment is minimal, targeted, and leaves no production footprint.
Do NOT use this skill for general research (silver:research) or for full
feature planning (silver:feature). A spike answers a specific yes/no or
comparison question. If the question is broad enough to require a PLAN.md,
use silver:plan instead.
Write .planning/spikes/<NNN>-<slug>/README.md for each experiment and
.planning/spikes/MANIFEST.md for the registry.
Each spike README must include:
# Spike: <title>
Hypothesis: <what you believe to be true>
Question: <the specific question this experiment answers>
Given/When/Then:
Given: <preconditions>
When: <action taken>
Then: <observable outcome that proves/disproves>
## Experiment
<code, commands, or configuration used>
## Result
VALIDATED | INVALIDATED | PARTIAL
Evidence: <what you observed — output, error, timing, diff>
## Decision
<what this means for the implementation approach>
MANIFEST.md must include:
# Spike Manifest
| ID | Slug | Question | Result | Date |
|----|------|----------|--------|------|
| Flag | Behaviour |
|---|---|
| (none) | Interactive intake — ask clarifying questions before running experiments |
--quick | Skip intake; use $ARGUMENTS text as the hypothesis directly |
--wrap-up | Package completed spike findings into a reusable project-local knowledge entry via silver:rem |
Display SILVER BULLET > SPIKE.
Understand the question. If not --quick, ask at most 3 targeted
clarifying questions:
Decompose into 2–5 experiments. Each experiment must:
Good experiments: prototype a call to an external API, benchmark two algorithms against a realistic data size, try a library's API to verify it handles the edge case, run a migration dry-run against a snapshot.
Bad experiments: "research approaches" with no code, multi-day
investigations, experiments that modify production data, broad codebase
exploration (use silver:scan for that).
Run experiments in order of risk. Start with the hypothesis most likely to short-circuit remaining experiments if invalidated.
Record each result immediately. Write the spike README before moving to the next experiment. Do not defer recording.
Synthesize. After all experiments complete, write a one-paragraph
synthesis in MANIFEST.md:
File deferred items. Any question raised that was out of scope for
this spike → silver:add.
Wrap-up (if --wrap-up). Invoke silver:rem with the key findings
so they survive context compaction and are available in future sessions.
Handoff. If the spike result feeds a planning decision, surface the
synthesis in CONTEXT.md or pass it as input to silver:clarify.
--wrap-up is specified, record only the decision and evidence —
never the throwaway code.The spike is complete only when:
MANIFEST.md contains a synthesis and a recommended next step;silver:add.A spike that ends in PARTIAL on all experiments without a recommended next step is not a completed spike — surface the ambiguity and ask the user how to proceed.
npx claudepluginhub alo-exp/silver-bullet --plugin silver-bulletGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.