docs(skill): add cleveragents-spec skill from docs/specification.md
New skill covering every architectural concept, entity, workflow, CLI
command, and design decision from docs/specification.md (47,181 lines
read in full). Explains WHAT the system is intended to build.
SKILL.md (1,282 lines) — 18 decision trees:
- 'What am I working on?' master routing tree
- 'What is a Plan?' — 4 phases, reversion rules, hierarchy, decision tree
- 'How does a plan run?' — step-by-step Action→Strategize→Execute→Apply
- 'What is a Decision?' — 10 types, data model, dual tree+DAG structure,
timing by phase, decision recording protocol
- 'How do I correct a plan?' — revert vs append modes, Strategize vs
Execute correction mechanics, affected subtree computation
- 'What is an Invariant?' — 4 scopes, precedence chain (plan>action>
project>global), non-overridable globals, Invariant Reconciliation Actor
- 'What is an Actor?' — LLM vs graph types, Jinja2+env-var preprocessing,
specialized roles (strategy/execution/estimation/invariant)
- 'What is a Tool?' — 4 sources, capability metadata, 4-stage lifecycle,
resource bindings and slots, anonymous tools, metadata overrides
- 'What is a Validation?' — Tool subtype, always read-only, required vs
informational modes, 3 attachment scopes, wrapping existing tools
- 'What is a Skill?' — composition patterns, includes, tool overrides
- 'What is a Resource?' — physical vs virtual, 34+ built-in types, DAG,
type inheritance, 5-sandbox strategies, 6-level execution env routing,
devcontainer auto-discovery and lazy activation
- 'What is a Project?' — resource linking, multi-project plans, context
config, execution environment
- 'Which automation profile applies?' — 8 built-in profiles, 11 flags,
Safety Profile, Automation Guard, Semantic Escalation, progressive trust
- 'How does naming work?' — namespace format, types, ULID vs name identity
- 'Which CLI command do I use?' — every command group with key flags
- 'What is the architecture?' — 4 layers, 2 deployment modes, A2A
protocol (full method routing, error codes, streaming), DI container
- 'What is the ACMS?' — UKO, CRP, 10-slot pipeline, hot/warm/cold tiers
- 'Which milestone am I in?' — v3.2.0–v3.8.0 status + cross-milestone invariants
- Key Numbers table (35 entries)
Reference files (1,974 lines across 9 files):
- plan-lifecycle: phase mechanics, decision tree schema, checkpoint triggers,
child plan execution modes, merge strategies, plan identity fields
- entities: data models for Plan, Decision, Action, Session, Invariant,
AutomationProfile, SafetyProfile, AutomationGuard, Namespace
- architecture: 4-layer diagram, deployment modes, complete A2A method
routing tables (standard + plan + registry + context + sync + health),
streaming events, authentication, error taxonomy, full tech stack
- automation-profiles: threshold table for all 8 built-in profiles, use
cases, Semantic Escalation algorithm, custom profile YAML
- actors-tools-skills: Actor/Tool/Validation/Skill YAML schemas with
complete annotated examples, Jinja2 filter reference, LSP integration
detail, LSPToolAdapter, actor context precedence
- resources: complete resource type hierarchy (all 34+ types), sandbox
strategies, type inheritance rules, execution environment routing,
devcontainer integration, CLI usage
- acms: UKO 4-layer ontology, CRP, 10-slot Context Assembly Pipeline with
per-slot component names, hot/warm/cold eviction rules, skeleton
compression, context view configuration
- milestones: v3.2.0–v3.8.0 deliverables, architectural constraints, and
definitions of done; cross-milestone quality gates and invariants
- cli-commands: complete CLI reference for all command groups with all
flags: plan, action, session, project, actor, skill, tool, validation,
resource, invariant, automation-profile, lsp, config, utility
ISSUES CLOSED: #0
freemo
5a24b5bdf2