Skillget

sdp

agent
Security Audit
Pass
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 17 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This agent provides a structured development protocol for AI-assisted workflows, guiding tools through a Discovery, Delivery, and Evidence pipeline. It generates verifiable proof of the work completed by the AI and can gate pull requests based on this evidence.

Security Assessment
Risk Rating: Low
The static code scan of 12 files found no dangerous patterns, no hardcoded secrets, and no dangerous permissions requested. However, the primary installation method relies on piping a remote shell script directly into the environment (`curl ... | sh`). While standard practice for many CLI tools, this execution method inherently bypasses manual code review before running the script on your machine. The tool integrates with system Git and CI environments to verify and merge code, so it will read local repository data, but it does not appear to access unauthorized sensitive files or make suspicious outbound network requests.

Quality Assessment
Risk Rating: Good
The project is highly active, with its last push occurring today. It is properly licensed under the permissive MIT license. While the community is currently small (17 GitHub stars), the documentation is thorough and outlines clear operational boundaries. The overall codebase appears clean, well-structured, and adequately maintained for early-stage adoption.

Verdict
Use with caution — the tool itself is safe, but you should manually review the `install.sh` script before executing it.
SUMMARY

Structured protocol for AI-assisted development. Discovery → Delivery → Evidence.

README.md

SDP: Structured Development Protocol

Protocol + evidence layer for AI agent workflows.

SDP gives your AI agents a structured process (Discovery → Delivery → Evidence) and produces proof of what they actually did. Works with Claude Code, Cursor, OpenCode, or anything that can read markdown.

Manifesto — what exists, what's coming, why evidence matters.

Quick Start

# Install (prompts, hooks, optional CLI)
curl -sSL https://raw.githubusercontent.com/fall-out-bug/sdp/main/install.sh | sh

# Or binary only
curl -sSL https://raw.githubusercontent.com/fall-out-bug/sdp/main/install.sh | sh -s -- --binary-only

# Or submodule
git submodule add https://github.com/fall-out-bug/sdp.git sdp

Skills load from sdp/.claude/skills/ (Claude) or sdp/.cursor/skills/ (Cursor).

If you embed SDP as a submodule inside another repo, use the public GitHub URL above as the source of truth. Do not point .gitmodules at a local sibling path such as ../sdp, or teammates and CI will drift onto commits nobody else can fetch.

First run:

sdp init --auto
@feature "Your feature"
@oneshot <feature-id>
@review <feature-id>
@deploy <feature-id>

→ 5-minute guide

What SDP Does

  1. Structures agent work — Intent → Plan → Execute → Verify → Review → Publish. Each phase has a contract.

  2. Produces evidence — JSON envelope with intent, plan, execution, verification, provenance (hash chain). Details.

  3. Gates PRs — sdp-evidence validate in CI. Incomplete evidence = blocked merge.

Core Workflow

Phase Command
Planning @vision "AI task manager" → @feature "Add auth"
Execution @oneshot <feature-id> or @build 00-001-01
Review @review <feature-id>
Deploy @deploy <feature-id>
Debug @debug, @hotfix, @bugfix

Done = @review APPROVED + @deploy completed.

Skills

Skill Purpose
@vision Strategic planning
@feature Feature planning (→ workstreams)
@oneshot Autonomous execution
@build Single workstream (TDD)
@review Multi-agent quality review
@deploy Merge to main
@debug / @hotfix / @bugfix Debug flows

Optional

CLI: sdp doctor, sdp status, sdp next, sdp guard activate, sdp log show, sdp demo

Beads: brew tap beads-dev/tap && brew install beads — task tracking for multi-session work.

Platform: Evidence layer uses flock — macOS/Linux only. Windows not supported.

Research Lab: We're exploring multi-persona adversarial review, self-improvement loops, cross-project federation, and telemetry-driven backlog generation in sdp_lab. Private for now — open an issue if you'd like to play with us.

Docs

File Content
QUICKSTART.md 5-minute getting started
MANIFESTO.md Vision, evidence, what exists
ROADMAP.md Where SDP is going
PROTOCOL.md Full specification
reference/ Principles, glossary, specs

License

MIT

Reviews (0)

No results found