Agentic SE Skills

Source-grounded agent skills inspired by the NASA Systems Engineering Handbook (SP-2016-6105 Rev 2), adapted for agentic software development. This is a skill pack and plugin—not NASA certification, and handbook text is context for design, not proof that an agent performed correctly.

What you get

Nine skills: one router plus eight direct skills. Each skill includes SKILL.md, workflow steps, and page-grounded references/source-grounding.md.

Skill	Use when
agentic-systems-engineering	Mixed SE intent; routes to internal playbooks (ConOps, requirements, V&V, decisions, assessment)
conops-builder	User scenarios, actors, nominal/off-nominal flows before coding
requirements-distiller	Turning vague intent into testable, traceable requirements
requirements-auditor	Ambiguity, contradictions, unverifiable or scope-creep requirements
integration-planner	Multi-component sequencing, merges, interfaces, integration risk
verification-planner	Tests, inspections, analyses, demos against stated requirements
validation-planner	Fit for intended use, user workflow, operational need
decision-analysis	Architecture/tool trade studies, criteria, decision records
technical-assessment-loop	Checkpoints: plan vs evidence, risk, readiness before continuing

Also included: NASA-SE codebase assessment rubric, extraction artifacts, eval prompts, and contract scripts for maintainers.

Quick start

Clone the repo, then pick your agent harness:

Harness	Install
Cursor	./scripts/install_plugin.sh then fully quit and reopen Cursor (verify 9 skills)
Codex Desktop	./scripts/install_plugin.sh or ./scripts/install_plugin.sh --codex (requires codex CLI)
Hermes Agent	./scripts/install_hermes_skills.sh (details)
Try without global install	Open this repo in Cursor; run ./scripts/sync_harness_skills.sh so .cursor/skills/ exposes all nine skills

git clone https://github.com/jon-devlapaz/agentic-se-skills.git
cd agentic-se-skills
./scripts/install_plugin.sh          # Cursor + Codex
# or
./scripts/install_hermes_skills.sh   # Hermes Agent

Example prompts

In agent chat (slash commands vary by harness):

• /agentic-systems-engineering — “We need a release plan for this feature; separate requirements from verification and validation.”
• /requirements-distiller — “Turn this vague feature request into testable requirements with acceptance criteria.”
• /verification-planner — “Given these requirements, what tests and evidence prove we built the right thing?”
• /validation-planner — “Does this implementation match how operators actually run the workflow?”

Important boundaries

• NASA handbook excerpts explain why a skill step exists; they do not prove your agent or codebase is correct.
• The codebase assessment rubric is not official NASA certification or endorsement.
• The router is a viable single activation point; it does not replace or outperform the eight direct skills in aggregate (see AGENTS.md).

Develop and contribute

Requirements: Python 3.11+

python3.11 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -e ".[dev]"
pytest -q
python scripts/check_all_contracts.py

See CONTRIBUTING.md for PR expectations. Agent/maintainer rules: AGENTS.md.

After editing skills under .agents/skills/, run ./scripts/sync_harness_skills.sh before release.

Deep dives

Topic	Doc
Cursor, Codex, Hermes install	docs/plugin-harnesses.md, docs/hermes.md
NASA source and PDF	docs/SOURCE_ATTRIBUTION.md
Extraction layout	extraction/ (pages, figure/table indexes)
Eval prompts	evals/
Architecture decisions	docs/adr/

Maintainers: regenerate extraction

The committed extraction uses pypdf in src/agentic_se_skills/extract_handbook.py. Regenerate only when the source PDF or grounding references change intentionally.

pip install -e .
python -m agentic_se_skills.extract_handbook \
  --pdf sources/nasa_systems_engineering_handbook_0.pdf \
  --out .

Rendered PNGs in the repo are a pruned subset of pages referenced by skills and indexes; a full 297-page render can be regenerated from the PDF when needed. Page-range iteration (faster):

python -m agentic_se_skills.extract_handbook \
  --pdf sources/nasa_systems_engineering_handbook_0.pdf \
  --out /tmp/agentic-se-sample \
  --start-page 43 \
  --end-page 46 \
  --no-render-pages

Note: Full page rendering uses macOS sips today. Linux maintainers can run contract tests and text extraction without regenerating renders unless they add an alternate render path.

Before release:

python scripts/check_all_contracts.py

License and sources

Repository code is MIT. NASA Systems Engineering Handbook material is attributed separately—see docs/SOURCE_ATTRIBUTION.md.