JGS Systems Engineering Knowledge Packs
Turn open NASA, DoD, NIST and FAA systems-engineering bodies of knowledge into on-demand Agent Skills, with a /se router to the right chapter.
- Source repo
- jgsystemsconsulting/jgs-se-knowledge-packs
- Stars
- ★ 89
- Last updated
- 1d ago
- License
- NOASSERTION
- Primary language
- Python
- FA score
- 47/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesClaude CodeCodex (Partial support)OpenAI API (Unsupported)
- Cost
- Free, no paid service needed
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A systems engineer on an aerospace or defence programme invokes
/se what are the EARS requirement patterns?to pull requirement-writing guidance mid-task instead of hunting through a PDF. - Not a fit if
- Teams that need the verbatim standard text, not reconstructed notes
- Users needing non-English pack content
- Teams wanting a zero-install web Q&A interface
- Source review
- 47/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
This is a curated catalogue of 74 knowledge-pack skills that reconstruct genuinely open systems-engineering sources — NASA SE Handbook, NASA NPR 7123.1D, NIST SP 800-160, MIL-STD-882E, FAA SEM, SEBoK and more — into Agent Skills format. Each pack ships an always-loaded SKILL.md index plus on-demand chapters/ with a glossary, patterns and a cheatsheet. Eight citation-only signposts map the paywalled standards landscape, and a /se orchestrator routes a question to one to four packs and answers with pack and chapter citations. Installation goes through the repo's install.py (default ~/.claude/skills/jgs-se-knowledge-packs/) or the Vercel skills CLI; packs are plain Markdown and need no MCP server, API key or licence tier at runtime. Licensing is split: MIT for tooling, each source's own terms for pack content.
The installer copies each pack into the host's skills directory and prints the pack list and target; on startup the host loads the always-loaded SKILL.md index. When a user types /sebok systems of systems or /sebok ch20, the agent follows the index to the matching chapters/ file plus glossary, patterns and cheatsheet, and answers from those notes. The /se orchestrator carries no source content: it holds a curated topic/agency/deliverable map, runs only when explicitly invoked, and produces a read plan over one to four packs with pack and chapter citations. install.py supports --dry-run, --flat, --list-agents and --agent <id>|all; native hosts (Claude Code, OpenClaw/Copilot CLI) get the full pack tree, while transform hosts (Codex, Gemini CLI, Cursor) get a single file with the SKILL.md index inlined as a prompt, TOML command or .mdc rule. Each pack carries PACK.yaml and a LICENSE reproducing its source's terms, and SKILLS.md plus catalog.json provide the machine-readable index.
- A systems engineer on an aerospace or defence programme invokes
/se what are the EARS requirement patterns?to pull requirement-writing guidance mid-task instead of hunting through a PDF. - A safety and risk review team uses the
mil-std-882,nasa-riskanddod-riopacks to compare MIL-STD-882E and NASA risk-acceptance positions chapter by chapter. - A team already using Claude Code, Copilot CLI or Amp installs the catalogue as project skills so the agent cites standard chapters rather than recalling them vaguely.
- A programme office checking acquisition compliance uses
se-standards-signpostoravionics-signpostto find who owns a given standard, whether it can be packaged, and where to get it. - Engineers in regulated domains (medical devices, nuclear, automotive) check
fda-med-device,nrc-nuclearorautomotive-signpostto see what is legally redistributable before building their own pack. - Trainers install the packs on student machines so newcomers can query the standards corpus through a single
/seentry point.
How do you install or deploy this agent?
Prerequisites: Python 3.9+, and a host that reads Agent Skills (Claude Code, GitHub Copilot CLI or Amp). No MCP server, API key or licence tier is needed.
Option 1 — let your coding agent install it: paste the README's install block into Claude Code or Cursor; it reads README.md, docs/skill-usage.md and CHANGELOG.md, then runs the installer.
Option 2 — clone and run the installer manually:
git clone https://github.com/jgsystemsconsulting/jgs-se-knowledge-packs.git
cd jgs-se-knowledge-packs
python install.py --dry-run # preview packs and target
python install.py # installs to ~/.claude/skills/jgs-se-knowledge-packs/Option 3 — pull packs with the Vercel skills CLI, no clone:
npx skills add jgsystemsconsulting/jgs-se-knowledge-packs -l # list available packs
npx skills add jgsystemsconsulting/jgs-se-knowledge-packs --skill sebok # install one pack
npx skills add jgsystemsconsulting/jgs-se-knowledge-packs --all # install the catalogueInstalling into other agents:
python install.py --list-agents # show every supported agent and its target
python install.py --agent openclaw # OpenClaw / Copilot CLI, native full pack
python install.py --agent codex # Codex / Gemini / Cursor, transformed SKILL.md index
python install.py --agent all # all user-global agents at onceRestart the host agent afterwards. The default target honours $CLAUDE_CONFIG_DIR; add --flat to place each pack directly under ~/.claude/skills/.
How do you use this agent?
After restarting the host, invoke a pack by slug or let /se pick one:
/sebok # load core frameworks for reference
/sebok systems of systems # route to the right chapter and answer
/sebok ch20 # load a specific chapter
/se what are the EARS requirement patterns? # orchestrator routes and cites
/se how do NASA and DoD differ on risk acceptance? # one plan, packs read togetherAnswers carry pack and chapter citations. Machine-readable indexes are SKILLS.md and catalog.json; per-pack provenance and terms are in NOTICE and packs/<slug>/LICENSE. Any redistribution must keep attribution and respect each pack's source licence (for example sebok is CC BY-NC-SA 3.0, jsp-376 is OGL v3.0).
What are this agent's strengths and limitations?
- Licence vetting happens before packaging: only public-domain or explicitly open sources qualify, and each pack's PACK.yaml, LICENSE and NOTICE entry records its source terms, so redistribution is traceable.
- Progressive disclosure by design: an always-loaded SKILL.md index plus on-demand chapters/, glossary, patterns and cheatsheet keeps whole handbooks out of the context window.
- Broad, file-level coverage: 74 packs span NASA, DoD, NIST, FAA, GAO, DOE, NRC, FDA, UK MoD and SEBoK, including MIL-STD-882E, MIL-HDBK-61B and JSP 936.
- Multi-host installer: install.py targets Claude Code, Copilot CLI and OpenClaw natively, and transforms packs into Codex prompts, Gemini TOML commands and Cursor .mdc rules.
- Plain-Markdown runtime with no MCP server, API key or licence tier required to load packs.
- Pack content is reconstructed reference notes, not the verbatim standard; exact clauses, tables and numeric limits still need checking against the source document.
- Transform hosts (Codex, Gemini CLI, Cursor) only receive the inlined SKILL.md index and lose the chapters/ progressive disclosure, so they are weaker than native installs.
- Pack licences differ (CC BY-NC-SA 3.0, OGL v3.0, per-document EASA grants); the repo MIT licence does not cover content, so commercial reuse needs per-pack review.
- No independent test evidence is published for the packs or installer, and at least one catalogue entry (mit-ocw-se) is still marked planned.
- Corpus is English-language systems-engineering material; there are no localised pack contents for non-English teams.
How does this agent compare with similar options?
Against pointing a retrieval tool at a PDF, this project vets sources by licence tier (Tier 1 public domain, Tier 2 open-licensed) first, maps each source to chapter offsets, and distributes the result as Agent Skills SKILL.md packs. For paywalled standards (ISO/IEC/IEEE, INCOSE, SAE, ECSS), it ships citation-only signposts that name the owner and acquisition path instead of reproducing content.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| JGS Systems Engineering Knowledge Packs This agent | 47 · Major gaps | Agent plugin / skillFree | ★ 89 | 1d ago | Python | Claude Code |
| Finding-Unknowns Skills | 76 · Good | Agent plugin / skillFree | ★ 363 | 1mo ago | Python | Claude Code · Claude.ai |
| Keep the Why | 75 · Good | Agent plugin / skillFree | ★ 233 | 1d ago | JavaScript | Codex · Claude Code |
| Project Cairn | 67 · Some gaps | Agent plugin / skillFree | ★ 253 | 14d ago | Shell | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The repo claims plain-Markdown skills plus stdlib Python tooling with no runtime server or credential handling; SECURITY.md logs accepted risks and states no pip/npm dependencies, so dependency surface is small (dependency_security 2). However install.py writes into the user home (~/.claude/skills etc.) with no documented overwrite, uninstall, or rollback path (rollback 0); installation is driven by a prompt block the user pastes to an agent, with no explicit confirmation gate (user_confirmation 1); data-flow and external-effect disclosure is thin (1 each). Source attribution is strong: NOTICE, per-pack LICENSE, PACK.yaml, and a two-layer licence model (source_attribution 2).
README, SKILLS.md, catalog.json, CHANGELOG and CI checks (version single-source, index counts, catalogue parity) show internal consistency is mechanically guarded (self_consistency 2); dependencies are only Python 3.9+ and a host that reads Agent Skills (dependency_availability 2). Failure messages and error handling are barely described outside CI FAIL lines, so failure_messages is 1.
Audience and scenarios are clear for systems engineers and coding agents, with multi-agent install paths and format transforms (audience_and_scenarios, environment_fit 2); SKILL.md declares per-pack thinness and /se has routing and gating rules (capability_boundaries 2). Trigger precision relies on description matching with no concrete trigger words or conflict handling shown, so trigger_precision is 1.
Information architecture is clear (README, docs, SKILLS.md, catalog.json, PACK-SPEC), install notes are thorough (2), naming has alias and stability notes (2), and licensing is layered and explicit (2). Examples and FAQ are sparse (1), known limitations are only briefly noted in skill-usage (1), CHANGELOG is referenced but not shown (versioning_changelog 1), and maintenance responsibility rests on a contact address while publisher identity is unverified (maintenance_responsibility 1).
Outputs carry pack and chapter citations plus a Sources block, aiding usability (output_usability 2); but packs are single-source reference oracles, limiting marginal value, and the 74-pack context-cost tradeoff is not quantified (marginal_value, cost_benefit 1 each).
Claims are nominally traceable to pack and chapter, but no actual citation samples or source-checking mechanism appear in the static files (claim_traceability 1); cross-source corroboration depends on the /se orchestrator design without demonstrated evidence (cross_source_corroboration 1); fact/inference separation is stated as a principle but lacks concrete examples (fact_inference_separation 1).
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- install.py writes into the user home directory with no documented overwrite, uninstall, or rollback path; back up and review install.py before installing.
- The README's agent-driven install prompt lets an agent run installation automatically; prefer running --dry-run manually and confirming the target path first.
- Publisher identity is unverified by FollowAgents; maintenance responsibility and update path rest on the in-repo contact address and should be independently confirmed.
- Packs are single-source reference oracles, not authoritative substitutes; re-check original standards for compliance or safety decisions.
- Licensing is layered (MIT tooling, per-source content); verify packs/<slug>/LICENSE and NOTICE per pack before commercial use.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v1.24.3 jgs-se-knowledge-packs 1.24.3
Release notes excerpt
Fixed - Landing page catalogue chip and website product YAML DoD list state MIL-HDBK-516D for the mil-hdbk-516 pack (public free-text had lagged on MIL-HDBK-516C after the pack rebuild). - Cursor marketplace exclusion story in docs/other-agents.md keys on PACK.yaml commercial_use: false and names the current set (sebok, easa-rules, med-device-eu), matching the eligible-set gate. Changed - Scaffold path and README "How a pack is built" retire offset-mapped / chapter-offset vocabulary; tooling/build_pack.py stamps the PACK-SPEC canonical build method string. …View on GitHub ↗ -
v1.24.2 jgs-se-knowledge-packs 1.24.2
Release notes excerpt
Fixed - docs/index.html hero title-card alt text states the current catalogue total (74 packs), replacing a stale 71 that predated the jsp-936 and fhwa-its-se registrations. The GitHub repository About description was corrected the same way (74 packs + 8 signposts + the /se orchestrator, 83 skills); repo tooling cannot reach the About field, so it was updated through the GitHub API. Changed - Rebuild nasa-de-acquisition onto NASA-HDBK-1004A, approved 2026-10-26 (8 chapters). Baseline 2020 edition pin retired. …View on GitHub ↗ -
v1.24.1 jgs-se-knowledge-packs 1.24.1
Release notes excerpt
Fixed - docs/LICENSING.md no longer teaches OMG SysML as a packable open licence. The Tier-2 section now states the SOURCE-VETTING.md exclusion (informational-use-only grant, two independent licence reads) and names omg-signpost as the only catalogue form for OMG specifications. - Author and operator docs restored to post-1.24.0 truth: inventory sentences state 74 content packs and 83 skills across SOURCE-FRESHNESS.md, PACK-SPEC.md, DISTRIBUTION.md, and CONTRIBUTING.md; PACK-SPEC.md gains a Build rules section (multi-PDF part pin, contents list wins over the TOC, verify-what …View on GitHub ↗