Instalar en Claude Code
Copiargit clone --depth 1 https://github.com/Filip-Podstavec/claude-leverage /tmp/adr-new && cp -r /tmp/adr-new/skills/adr-new ~/.claude/skills/adr-newDespués abre una sesión nueva de Claude Code; el skill carga automáticamente.
Definición
SKILL.md
# /adr-new
## What it does
Creates a new numbered ADR file in `docs/adr/` and adds an entry to the
index. ADRs are the lightweight memory of *why* the architecture looks
the way it does — without them, six months from now nobody (human or
agent) remembers why we chose A instead of B.
The convention is documented in
[`docs/adr/README.md`](../../docs/adr/README.md): MADR-flavored,
numbered, immutable status (proposed → accepted → deprecated /
superseded).
## Workflow
1. **Find target dir.** Default `docs/adr/` from the current repo root
(via `git rev-parse --show-toplevel`). If no `docs/adr/` exists,
ask the user: "create it now?". If yes, also copy
`templates/adr-template.md` if shipped (the plugin ships its own
template under `docs/adr/template.md` — useful as a reference even
when the target repo doesn't have one yet).
2. **Pick next number.** List existing `docs/adr/NNNN-*.md` files,
parse the leading number, take the max + 1. Pad to 4 digits.
3. **Get the title.**
- If `$ARGUMENTS` has a positional, use it as the title.
- If `--noninteractive`, fail if no title was given (we won't
hallucinate one).
- Otherwise, prompt the user.
- Convert to kebab-case for the filename:
`0042-replace-polling-with-webhooks.md`.
4. **Ask for context** (skip in `--noninteractive`):
- One-line summary of the problem.
- The chosen option (one sentence, stated as fact in present
tense).
- 1–3 alternatives considered (one line each).
- Whether to mark `status: proposed` (default) or `accepted` (if
the team already decided in conversation and just needs the
record).
5. **Generate the file** at `docs/adr/<NNNN>-<kebab-title>.md` using
the MADR template (the plugin ships
[`docs/adr/template.md`](../../docs/adr/template.md) for reference).
Substitute:
- `NNNN` → resolved number
- `Title` → user-provided title
- `YYYY-MM-DD` → today (`date +%Y-%m-%d`)
- `Status` → resolved value
- Context / Decision / Alternatives → user input or
`<TODO: fill in>` placeholders if user skipped them
6. **Update the index.** Append the new ADR to the bulleted list in
`docs/adr/README.md` (or create it if it doesn't exist). Preserve
the existing index style:
`- [NNNN — Title](NNNN-kebab.md)`
7. **Report.** Print the new file path and remind the user that ADRs
are immutable once `accepted` — future revisions are *new* ADRs
superseding the old one, not edits.
## Hard rules
- **Numbers are immutable.** Once shipped, an ADR's number never
changes. If you regenerate the file (because of typos in the title
before commit), keep the number.
- **Status is immutable once `accepted`.** Edits to an `accepted` ADR
should be limited to typos and link fixes. Substantive changes ship
as a new ADR that explicitly supersedes the old one
(`Status: superseded by NNNN`).
- **The index is the source of truth for "what ADRs exist."** Always
update it; never let it drift from the directory contents.
- **Never overwrite an existing ADR.** If the kebab-title collides
with an existing file, ask the user for a different title.
## Tunables
- `--status accepted` — start the ADR as `accepted` rather than
`proposed`. Use when the decision is already settled and you're just
recording it.
- `--noninteractive` — skip prompts; requires a title positional and
fills the context/decision/alternatives with `<TODO>` placeholders.
- `--dry-run` — print what would be written, write nothing.
## When to write an ADR (and when not to)
**Write an ADR for:**
- Architectural tradeoffs likely to be re-litigated (e.g., "no
embedding RAG", "AGENTS.md canonical, CLAUDE.md import").
- Decisions that depend on external constraints (a vendor, an open
issue, a benchmark result).
- Decisions where alternatives are plausible enough that a future
agent will propose them.
**Don't write an ADR for:**
- Implementation choices the code itself shows (variable names,
function structure).
- One-off tactical fixes.
- Things covered by an obvious convention (test naming, lint config).
If you're not sure: write it. Three sentences in `docs/adr/0042.md`
is cheap; re-arguing the same point six months later is expensive.
## What this skill does NOT do
- **Doesn't make the decision.** It records a decision you've already
made. If you don't yet know the decision, you want a `docs/specs/`
document (or `superpowers:brainstorming`), not an ADR.
- **Doesn't read the conversation context.** You tell it the
decision in your own words; the skill just structures the file.
- **Doesn't supersede or modify existing ADRs.** Modification is a
manual edit + new ADR with `Status: superseded by NNNN`.Del mismo repositorio
flaky-test-isolatorSubagent
USE WHEN a test intermittently fails on unchanged code. Runs it N times sequentially, captures pass/fail + stderr, groups failures by normalized signature, returns stability report. Read-only — never modifies code or installs deps. For statistical signal across runs, not one-shot diagnosis.
security-reviewerSubagent
USE BEFORE committing security-sensitive changes (auth, crypto, routes, templates, secrets). Audits current diff for OWASP-Top-10 patterns + deps typosquatting. Read-only. Returns Critical / Important / Nice schema with file:line. Model review — not a Semgrep/CodeQL replacement.
flaky-testSlash Command
Diagnose a flaky test by running it N times. Delegates to flaky-test-isolator subagent — N runs, signature-grouped failures, stability report. Does NOT fix the test.
arch-mapSkill
>
codex-sandboxSkill
>
conventions-initSkill
>
explain-diffSkill
>
glossary-initSkill
>