| Agent Skills |
The cross-vendor SKILL.md standard for prose-described skills. AISP is complementary to it (executable + governable), and authoring/distribution tools SHOULD include a thin same-folder SKILL.md sidecar by default for interop unless the author opts out. |
| AIAP |
AI Application Protocol — a sibling protocol in the AIXP family, not part of AISP's stack. aiap.dev. |
| AISOP |
AI Standard Operating Protocol — the execution language a skill body is written in (.aisop.json, Mermaid/JSON flow, the sys.* family); AISP's only base. aisop.dev. |
| AISP |
AI Skill Protocol — this protocol; the independent skill-package layer built on AISOP only. aisp.dev. |
aisp.aisop.json |
The fixed skill-file name. Structurally a legal AISOP V1.0.0 program (a 2-message array). One per skill folder. |
aisp_contract |
The skill contract — a real JSON object at user.content.aisp_contract: profile, invocation, non_negotiable, discovery, risk_level, resources. Tier S. |
aisp_list.json |
The discovery index cache: one row per skill (id, name, summary, path, category, tags, when_to_use, risk_level). Derived from the folders. |
aisp_list.py |
The zero-dependency discovery script: globs *_aisp/aisp.aisop.json, reads each aisp_contract, prints a list, --json regenerates the index. Running it = executing code. |
_aisp suffix |
The reserved skill-folder suffix. The folder name == id and ends with _aisp (mirroring the convention the sibling AIAP protocol uses for _aiap). Makes a skill folder self-identifying. |
| Axiom 0 |
"Human Sovereignty and Wellbeing." The immutable alignment principle inherited from HSAW; sys.io.confirm is forced-blocking. |
| Discovery |
The runtime-independent process of finding skills: read the index, run the script, or glob the folders. |
enforced_by |
The binding from a non_negotiable rule to a real mechanism: <node>.stepN:<mechanism> | aisop.main | tools. The step segment must be a numeric execution step such as step1; metadata fields such as step_note do not count. The policy-as-code layer. |
| Sidecar bridge |
A default-generated but core-optional same-folder thin SKILL.md that only guides loading/running aisp.aisop.json. Never copies logic; deletable without breaking the skill under an AISP/AISOP runtime. |
flow_format |
The skill's flow representation: "mermaid" | "jsonflow" | "hybrid". |
| Folder-per-skill |
The structural rule: one skill = one self-contained folder aisp/<id>_aisp/, zippable / hashable as a registry unit. |
functions |
The per-node execution steps (Tier A). Holds role, principles, knowledge, failure handling, and structured output. |
instruction |
The user-message directive, fixed as "STRICTLY OBEY aisp_contract; its non_negotiable rules are inviolable; then RUN aisop.main". |
invocation |
The triggering metadata (mode, when_to_use, when_not_to_use) — decided before execution. |
execute_mode |
A per-node dispatch field inherited from AISOP (§5.2.8): "inline" (default current context for short/simple nodes) | "agent" (independent sub-agent for isolation, source gathering, complex validation, generation, or high-impact decisions). If omitted, runtime falls back to inline and static validation warns. >10 numeric stepN execution steps is a review heuristic for non-agent nodes, not a restriction on shorter agent nodes. Honored by R7; a dispatch attribute, not an enforced_by mechanism. |
loading_mode |
The AISP skill loading strategy. AISP skills fix this to "node" for progressive disclosure. |
non_negotiable |
The red lines — an array of {rule, enforced_by}. The declaration ↔ enforcement binding that differentiates AISP from prose skills. |
| Open-world validation |
AISOP's tolerance of unknown keys and non-gating on the protocol value, which lets AISP add aisp_contract / license and declare protocol: "AISP V1.0.0" (R4). |
| policy-as-code |
A rule enforced deterministically by the runtime via enforced_by → sys.*, graph topology, or the tool allow-list. Hard. |
| policy-as-prompt |
A rule stated in natural language that the model is asked to follow. Soft. |
profile |
The contract schema marker, "aisp.skill.v1"; a runtime detects an AISP skill by the aisp.skill. prefix. |
| Progressive disclosure |
Loading metadata first (L1 index), then the full skill on a hit (L2), then resources on demand (L3). |
protocol |
The authorship marker, fixed "AISP V1.0.0"; the file is still structurally an AISOP program. |
resources |
The resource inventory (id, path, kind, mode, …) in the contract. Inventory, not usage; nodes use resources via sys.io.read / sys.run. |
risk_level |
A single risk grade: low | medium | high | critical. |
| Skill file |
The aisp.aisop.json of a skill. |
| Skill folder |
An aisp/<id>_aisp/ directory. |
| Skill package |
The aisp/<id>_aisp/ folder as a whole — the unit of distribution. |
sys.* |
AISOP's deterministic system-call family (sys.assert, sys.io.confirm, sys.io.read, sys.run, sys.llm.json, sys.code.exec, …) handled by the runtime, not by LLM reasoning. The source of all hard guarantees. |
| Tier A |
The execution tier — aisop.main + functions + sys.* + tools / params. Heavy. |
| Tier S |
The contract tier — user.content.aisp_contract. Lean metadata only. |
| Trust (not self-declared) |
A skill MUST NOT claim trusted / verified / safe; trust is judged by the runtime / registry / user / scanner (M6 / SE6). |
_shared/ |
A cross-skill shared-resources directory. Not a skill folder; no _aisp suffix; reached via scope:shared. |