Skip to content
Back to skills

Sdd Evidence

ASecurity

Discover, build, verify, and optionally export (HTML, optional PDF) the evidence manifest of one SDD Composy task in evidence_required, routing it toward review_required. Use when acceptance evidence must be assembled or checked — 'juntar as evidências da task', 'gerar o relatório de evidências', 'verify the evidence manifest'. Do NOT use to run tests (sdd-qa), review code (sdd-review), or inspect the trace (sdd-trace).

  • 3 stars
  • 0 votes
  • 2 copies
  • 3 views
  • Added September 22, 2026
research

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned September 22, 2026

npx -y skills add pwdev-solucoes/pwdev-claude-marketplace --skill sdd-evidence --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Sdd Evidence?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Sdd Evidence
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/pwdev-solucoes-sdd-evidence/badge)](https://www.skillsdirectory.com/skills/pwdev-solucoes-sdd-evidence)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: sdd-evidence
description: >
  Discover, build, verify, and optionally export (HTML, optional PDF) the evidence
  manifest of one SDD Composy task in evidence_required, routing it toward
  review_required. Use when acceptance evidence must be assembled or checked —
  'juntar as evidências da task', 'gerar o relatório de evidências', 'verify the
  evidence manifest'. Do NOT use to run tests (sdd-qa), review code (sdd-review), or
  inspect the trace (sdd-trace).
metadata:
  version: 0.1.0
---

# SDD Evidence

Build, verify, and optionally export the portable evidence manifest of one task with the
bundled `scripts/sdd_evidence.py` helper.

Language: when an operation emits human-facing summaries, run `scripts/sdd_language.py <repo-root>` and use the persisted language; on `not_initialized`, return it with `next_action: run_init`. Localization rules: `references/language.md`.

## Operations

1. `sdd_evidence.py discover <root>` — a deterministic, read-only inventory of the existing
   task evidence under the confined evidence root, before creating anything.
2. `sdd_evidence.py build <input.json> <root> <manifest>` — rebuild the manifest, preserving
   the source files and recording relative paths, SHA-256 digests, known enums, and sanitized
   content. Never mutate existing evidence while rebuilding.
3. `sdd_evidence.py verify <manifest> <root>` — required before routing a task from
   `evidence_required` to `review_required`. The task engine remains the authority and rejects
   missing, stale, or unapproved evidence.
4. `sdd_evidence.py export <manifest> <root> <output> [--workspace-root <repo>]` — HTML report.
   With the evidence root at `tasks/prd-<slug>/evidences/`, the output may be anywhere in that
   PRD bundle (normally `tasks/prd-<slug>/evidence-report.html`) and the workspace language is
   read from the repository root without `--workspace-root`. PDF is optional: `--pdf`, only when
   explicitly requested, writes `<output>.pdf` beside the HTML with a headless Chromium-family
   browser (`SDD_PDF_BROWSER` overrides discovery). Screenshots are embedded in the report; a
   missing or non-image screenshot fails, and without a browser the export fails before writing.

Never infer approval, bypass lifecycle guards, or use paths outside the task root. Return the
manifest and verification result unchanged to the caller.

## Read when

- `references/artifacts.md` ("Synchronization and evidence") — building a manifest, or a path,
  digest, or enum is rejected.

Safety: Do not commit, push, or publish. Do not read or expose `.env`, credentials, tokens, private keys, certificates, or fleet environment files. Full contract: `references/safety.md`.

Files in this skill

  • SKILL.md2.7 KB
  • agents/openai.yaml199 B

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.