DevKit — Workflow Diagrams

Visual workflows for the kit: core skills (hi-craft, hi-fix, hi-plan), the discovery-to-documentation pipelines, and the cross-skill integration graph. Catalog and invocation syntax live in README.md. Mode-selection rationale and delegation semantics: dev-shared/delegation-contract.md.

0. Skill Roles

0.1 Orchestrators

Skills that own global state, stage ordering, and publication. They are the only writers of checkpoint/manifest state.

Skill	Owns	Calls
`hi-maintenance-docs`	Complete multi-domain documentation pack (9 stages)	`hi-repo-recon`, `hi-tech-build-audit`, `hi-module-summary-report`, `hi-usecase-discovery`, `hi-api-contract-discovery`, `hi-data-model-discovery`, `hi-command-spec-discovery`, `hi-behavior-modeling`, `hi-cross-artifact-validation`, `hi-document-refinement`, `hi-document-review-resolution`
`hi-document-lifecycle`	One scoped domain, generate → clarify/refine → review/update with pause/resume	exactly one generator (`hi-module-summary-report`, `hi-usecase-discovery`, `hi-behavior-modeling`, `hi-api-contract-discovery`, `hi-data-model-discovery`, `hi-command-spec-discovery`) plus `hi-document-refinement` and `hi-document-review-resolution`
`hi-repowiki`	Repository wiki in `/repowiki/` (generate, update, plan-only)	`hi-repo-recon`, `hi-tech-build-audit`, `hi-module-summary-report`, applicable discovery leaves

0.2 Core workflow skills

Skill	Role
`hi-plan`	Implementation plans, architecture roadmaps, phased plans, plan validation
`hi-craft`	Feature implementation: plan → implement → test → finalize
`hi-fix`	Issue resolution across bugs, tests, CI, types, lint, UI, runtime
`hi-debug`	Root-cause debugging of failures, logs, CI, databases, performance

0.3 Leaf skills (called by main skills)

Skill	Called by	Purpose
`hi-codebase-research-explorer`	fix, debug, ad-hoc	Parallel file discovery and external research
`hi-repository-search`	plan, debug, ad-hoc evidence queries	Traceable repository evidence bundle
`hi-knows`	ad-hoc (no skill declares it as a leaf)	Knowledge retrieval from git, MCP, memory files
`hi-docs-seeker`	plan, debug, craft (full)	Authoritative external documentation lookup
`hi-log`	craft, plan, finalization flows	Session logging
`hi-sequential-thinking`	plan, ad-hoc reasoning requests	Step-by-step reasoning with revision
`hi-problem-solving`	fix, debug	Stuck-point techniques
`hi-project-organization`	any file-producing workflow	Output paths, names, layout
`hi-repo-recon`	maintenance-docs, repowiki, use-case discovery	Module inventory, entry points, key functions
`hi-tech-build-audit`	maintenance-docs, repowiki	Stack, build, CI/CD, deployment audit. `hi-module-summary-report` consumes its artifacts but must not invoke it.
`hi-module-summary-report`	maintenance-docs, repowiki, document-lifecycle	Architecture summary + risk synthesis
`hi-usecase-discovery`, `hi-behavior-modeling`, `hi-api-contract-discovery`, `hi-data-model-discovery`, `hi-command-spec-discovery`	maintenance-docs, document-lifecycle	Draft-only discovery artifacts
`hi-cross-artifact-validation`	maintenance-docs	Read-only cross-domain consistency validation
`hi-document-refinement`, `hi-document-review-resolution`	maintenance-docs, document-lifecycle	Semantic refinement and review-comment resolution

0.4 Analysis and review skills

Skill	Trigger
`hi-scenario`	Edge cases and test scenarios across 12 dimensions
`hi-brainstorm`	Ambiguous or consequential decisions: competing options, weighted scoring
`hi-predict`	Pre-analysis with GO/CAUTION/STOP verdict before risky changes
`hi-security`	STRIDE + OWASP audit, optional iterative auto-fix

0.5 Standalone tool and integration skills

Skill	Type	Purpose
`hi-repository-search`	Code + doc exploration	Search ingested repos via code graph (Qdrant/Neo4j/FalkorDb) + document graph RAG. Semantic search, call-graph tracing, dependency analysis.
`hi-cortex`	Harness infrastructure	Install and operate Cortex Harness: config, code/doc sync, database infrastructure, MCP lifecycle for use by other skills.
`hi-chrome-devtools`	Browser automation	Browser automation via Puppeteer CLI with persistent sessions. Screenshots, performance, network, scraping, form automation, auth, debugging.
`slide-kit`	Deliverable generation	`.pptx` decks in the Modernization corporate style, brand-neutral. No skill calls it; it answers direct deck requests.
`typesafe-ai`	AI feature building	TypeSafe System One models (Jev): typed judgments (Choice/Score/Noul) as programming primitives. The optional judgment backend behind dev-shared/jev-contract.md.
`hi-reverse`	Reverse engineering	C/C++ → use-case, module, and C++-to-Java migration artifact packages; self-contained pipeline with package gates.
`dev-shared`	Support package	Cross-skill contracts and role briefs. Never activated directly.

1. Core Pipelines

1.1 `hi-craft` — Feature Implementation

Mode Matrix

Mode	Research	Plan	Review	Test	Finalize
`fast` (default)	skip	inline `hi-plan --fast`	skip	run	commit + log
`full`	yes (mode-dependent)	yes	MUST	run	commit + log + review
`review`	skip	inline	MUST	run	commit + log + review
`auto`	skip	inline	auto-pass	run	commit + log
`no-test`	skip	inline	skip	skip	commit + log
`code` (path to plan)	skip	—	optional	run	commit + log

The plan gate applies in every mode: a plan is required unless the user explicitly says "just code it" / "skip planning".

graph LR
    A["User request"] --> B["hi-craft"]
    B --> C["Plan exists?<br/>hi-plan --fast if missing"]
    C --> D{"Parallel mode?"}
    D -->|no| E["Implement directly"]
    D -->|yes| P["Launch implementer<br/>per phase"]
    P --> E
    E --> F["Run proportionate checks"]
    F -->|pass| G["Finalize<br/>commit + hi-log"]
    F -->|fail x1-x2| E
    F -->|fail x3+| H["Spawn hi-fix<br/>deep debug"]

    classDef primary fill:#c8e6c9,stroke:#1b5e20,color:#000
    classDef delegated fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef gate fill:#ffe0b2,stroke:#e65100,color:#000
    class B,E,G primary
    class P,H delegated
    class C,D,F gate

1.2 `hi-fix` — Issue Resolution

Mode	Scope	Repository evidence	Delegation
default	1 file, type/lint, or clear issue	Direct inspection	1 agent is enough
`--standard`	2-5 files	Full explorer and diagnosis	`hi-codebase-research-explorer` or `2-3` parallel agents when useful
`--deep`	5+ files or architectural impact	Parallel explorer, diagnosis, research	Parallel agents for exploration/research
`--parallel`	2+ independent issues	Separate task tree per issue	Spawn `implementer` per issue
`--review`	Any scope	Mode-dependent	Human review at gates

graph LR
    A["Capture symptom<br/>error / log / failing check"] --> B["Explore affected area"]
    B --> P{"Standard / deep / parallel?"}
    P -->|no| C["Diagnose<br/>symptom -> cause -> root cause"]
    P -->|yes| X["hi-codebase-research-explorer or parallel agents<br/>locate evidence"]
    X --> C
    C --> D["Fix root cause<br/>minimal change"]
    D --> E["Verify"]
    E -->|pass| F["Finalize<br/>evidence + change + verification"]
    E -->|fail x1-x2| C
    E -->|fail x3| G["Stop and question architecture"]

    classDef step fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef gate fill:#ffe0b2,stroke:#e65100,color:#000
    classDef fail fill:#ffcdd2,stroke:#b71c1c,color:#000
    class A,B,C,D,F,X step
    class P,E gate
    class G fail

1.3 `hi-plan` — Implementation Planning

Mode	Research	Delegation	Review
default / `--fast`	Skip formal research; read docs or scan only if needed	None	None
`--full`	Research enabled	1 researcher	Optional red-team and validation
`--hard`	Deep research and impact analysis	2 researchers	Required red-team
`--parallel`	Parallel-track research	2 researchers	Required red-team
`--two`	Competing approaches	2+ researchers	Review after selection

graph LR
    A["Scan active plans"] --> B["Gather evidence<br/>hi-repository-search"]
    B --> R{"Research mode?"}
    R -->|fast| C["Define assumptions<br/>dependencies / gaps"]
    R -->|full / hard / parallel / two| S["Spawn researcher agents<br/>by selected mode"]
    S --> C
    C --> D["Write plan.md<br/>and phase files"]
    D --> E{"3+ phases?"}
    E -->|yes| F["Hydrate tasks"]
    E -->|no| G["Skip task overhead"]
    F --> H["Return plan path<br/>and execution command"]
    G --> H

    classDef step fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef evidence fill:#e1bee7,stroke:#6a1b9a,color:#000
    classDef delegated fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef gate fill:#ffe0b2,stroke:#e65100,color:#000
    class A,C,D,F,G,H step
    class B evidence
    class S delegated
    class R,E gate

Subcommands: /hi-plan archive (archive plans + log), /hi-plan red-team {path} (adversarial review), /hi-plan validate {path} (critical-questions interview).

graph LR
    Main[hi-plan] --> Default[default<br/>fast mode]
    Main --> Full[--full<br/>full flow]
    Main --> Hard[--hard<br/>2 researchers + red team]
    Main --> Parallel[--parallel<br/>2 researchers + red team]
    Main --> Two[--two<br/>2+ researchers,<br/>select after]
    Main --> NoTasks[--no-tasks<br/>skip task hydration]
    Main --> Archive[/hi-plan archive<br/>Archive plans + log/]
    Main --> RedTeam[/hi-plan red-team path<br/>Adversarial review/]
    Main --> Validate[/hi-plan validate path<br/>Critical questions interview/]

    classDef mode fill:#e3f2fd,stroke:#0d47a1,color:#000
    classDef sub fill:#c8e6c9,stroke:#1b5e20,color:#000
    class Default,Full,Hard,Parallel,Two,NoTasks mode
    class Archive,RedTeam,Validate sub

2. Discovery → Validation → Documentation

2.1 Discovery Artifact Pipeline

Five draft-only leaf skills produce evidence-backed artifacts for one module or scope; hi-cross-artifact-validation then checks consistency across them. Leaves never publish — the orchestrator owns refinement, review, and publication.

graph TD
    Scope["Scope evidence<br/>hi-repo-recon + hi-tech-build-audit"] --> UC["hi-usecase-discovery<br/>actors, rules, flows, outcomes"]
    Scope --> API["hi-api-contract-discovery<br/>OpenAPI/AsyncAPI drafts"]
    Scope --> DM["hi-data-model-discovery<br/>ERD, entities, lifecycle"]
    Scope --> CMD["hi-command-spec-discovery<br/>APDU/TLV, opcodes, status words"]
    UC --> BM["hi-behavior-modeling<br/>sequence, activity, state machine<br/>(requires a UC/entry-point anchor)"]
    UC --> XV["hi-cross-artifact-validation<br/>read-only, needs 2+ domains"]
    API --> XV
    DM --> XV
    CMD --> XV
    BM --> XV
    XV --> REF["hi-document-refinement<br/>analyze then apply"]
    REF --> REV["hi-document-review-resolution<br/>one disposition per comment"]

    classDef leaf fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef validate fill:#e1bee7,stroke:#6a1b9a,color:#000
    classDef gate fill:#ffe0b2,stroke:#e65100,color:#000
    class UC,API,DM,CMD,BM leaf
    class XV validate
    class Scope,REF,REV gate

Boundaries that keep this pipeline honest:

Each leaf is draft-only and states its own exclusion list (e.g. hi-command-spec-discovery is not for business scenarios or API contracts; hi-behavior-modeling is not for use-case catalogs).
hi-cross-artifact-validation is read-only and blocks when fewer than two domain manifests exist.
hi-document-refinement finds ambiguity and applies validated answers to an isolated candidate revision; it never self-publishes and never invents answers.
hi-document-review-resolution assigns exactly one disposition per comment (accepted, rejected, needs_information, duplicate, stale, out_of_scope, unsupported_by_evidence). Comment resolution and publication approval are separate actions.

2.2 Documentation Lifecycle

Two orchestrators, one stage contract. hi-maintenance-docs runs the full pack; hi-document-lifecycle runs a single domain with the same leaf skills.

graph TD
    Start(["Documentation request"]) --> Shape{"Breadth?"}
    Shape -->|"complete multi-domain pack"| MD["hi-maintenance-docs<br/>9 stages, own checkpoint + manifest"]
    Shape -->|"one scoped domain"| DL["hi-document-lifecycle<br/>generate → clarify/refine → review/update"]

    MD --> S1["1. hi-repo-recon + hi-tech-build-audit"]
    S1 --> S2["2. hi-module-summary-report"]
    S2 --> S3["3. hi-usecase-discovery"]
    S3 --> S4["4. hi-api-contract-discovery + hi-data-model-discovery + hi-command-spec-discovery"]
    S4 --> S5["5. hi-behavior-modeling<br/>after an anchor exists"]
    S5 --> S6["6. hi-cross-artifact-validation (optional)"]
    S6 --> S7["7. internal synthesis<br/>draft pack"]
    S7 --> S8["8. hi-document-refinement"]
    S8 --> S9["9. hi-document-review-resolution"]

    DL --> G1["Phase 1: one generator<br/>by domain routing"]
    G1 --> G2["Phase 2: hi-document-refinement"]
    G2 --> G3["Phase 3: hi-document-review-resolution"]

    S9 --> Approve{"review_required?<br/>explicit approval artifact"}
    G3 --> Approve
    Approve -->|approved| Publish["Publish index + current/"]
    Approve -->|blocking questions| Wait["Persist, stop in waiting_for_input<br/>resume revalidates hashes"]

    classDef orch fill:#c8e6c9,stroke:#1b5e20,color:#000
    classDef stage fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef gate fill:#ffe0b2,stroke:#e65100,color:#000
    class MD,DL orch
    class S1,S2,S3,S4,S5,S7,G1,G2 stage
    class S6,S8,S9,G3,Approve,Wait,Publish gate

2.3 Repository Wiki Pipeline

hi-repowiki is a third orchestrator with its own entry conditions: it generates or updates a /repowiki/ tree, supports generate / update / plan-only modes, en / zh output, and --resume.

graph LR
    R["hi-repowiki<br/>mode: generate/update/plan-only"] --> RC["hi-repo-recon (deep)"]
    R --> TB["hi-tech-build-audit (deep)"]
    R --> LF["applicable discovery leaves"]
    RC --> MS["hi-module-summary-report"]
    TB --> MS
    MS --> W["/repowiki/<br/>architecture, install, config,<br/>CLI/build, APIs, testing, deployment, modules"]
    LF --> W
    R -.->|"unsupported facet"| U["unknown / not_applicable<br/>never a fabricated page"]

    classDef orch fill:#c8e6c9,stroke:#1b5e20,color:#000
    classDef step fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef out fill:#fff9c4,stroke:#f57f17,color:#000
    class R orch
    class RC,TB,LF,MS step
    class W,U out

2.4 `hi-reverse` — self-contained reverse-engineering pipeline

hi-reverse does not join the documentation pipeline; it runs its own gated sequence over C/C++ sources and emits migration-ready packages.

graph LR
    C["CATALOG"] --> P["PROFILE"]
    P --> PL["PLAN"]
    PL --> D["DISCOVERY"]
    D --> T["TRACING"]
    T --> A["UC / SEQ / CLASS / ACTIVITY"]
    A --> M["MODULE MAP"]
    M --> I["OVERVIEW / ENTRYPOINTS / INTERFACES"]
    I --> S["STATE / DATA / RULES / ERRORS / CONCURRENCY"]
    S --> MG["MIGRATION PARITY / MAPPINGS / WAVES / RISKS / TESTS"]
    MG --> RV["REVIEW"]
    RV --> V["VALIDATION"]
    V --> G["PACKAGE GATE"]
    G --> ME["METRICS"]

    classDef step fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef gate fill:#ffe0b2,stroke:#e65100,color:#000
    class C,P,PL,D,T,A,M,I,S,MG,RV,V,ME step
    class G gate

3. Cross-skill Integration

graph TD
    User(["User request"]) --> Detect{"Intent"}

    Detect -->|"feature / implementation"| Craft["hi-craft"]
    Detect -->|"bug / failure"| Fix["hi-fix"]
    Detect -->|"plan / architecture"| Plan["hi-plan"]
    Detect -->|"repo question / impact"| RepoSearch["hi-repository-search"]
    Detect -->|"debug investigation"| Debug["hi-debug"]
    Detect -->|"security audit"| Security["hi-security"]
    Detect -->|"edge cases / tests"| Scenario["hi-scenario"]
    Detect -->|"ambiguous decision"| Brainstorm["hi-brainstorm"]
    Detect -->|"risky change pre-check"| Predict["hi-predict"]
    Detect -->|"browser task"| Chrome["hi-chrome-devtools"]
    Detect -->|"documentation pack / wiki"| Docs["hi-maintenance-docs / hi-repowiki"]
    Detect -->|"C/C++ reverse engineering"| Reverse["hi-reverse"]
    Detect -->|"deck / pptx"| Slide["slide-kit"]

    Craft -->|"plan needed"| Plan
    Craft -->|"verification fails repeatedly"| Fix
    Craft -->|"final record"| Log

    Fix -->|"root cause difficult"| Debug
    Fix -->|"2+ failed hypotheses"| PS["hi-problem-solving"]
    Fix -->|"final record"| Log

    Plan -->|"code/doc evidence"| RepoSearch
    Plan -->|"complex reasoning"| ST["hi-sequential-thinking"]
    Plan -->|"docs for a library/API"| DocsSeeker["hi-docs-seeker"]
    Plan -->|"outputs / naming"| Org["hi-project-organization"]
    Plan -->|"archive / final record"| Log

    Debug -->|"find files / external research"| Explorer["hi-codebase-research-explorer"]
    Debug -->|"repo evidence"| RepoSearch
    Debug -->|"stuck"| PS

    Docs -->|"structure evidence"| Recon["hi-repo-recon"]
    Docs -->|"build evidence"| Audit["hi-tech-build-audit"]
    Docs -->|"synthesis"| Summary["hi-module-summary-report"]
    Docs -->|"draft artifacts"| Leaves["hi-usecase-discovery / hi-api-contract-discovery / hi-data-model-discovery / hi-command-spec-discovery / hi-behavior-modeling"]
    Docs -->|"cross-domain check"| XV["hi-cross-artifact-validation"]
    Docs -->|"refine / resolve comments"| DLC["hi-document-refinement / hi-document-review-resolution"]
    Docs -->|"final record"| Log

    Cortex["hi-cortex"] -->|"healthy MCPs"| RepoSearch

    classDef primary fill:#c8e6c9,stroke:#1b5e20,color:#000
    classDef evidence fill:#e1bee7,stroke:#6a1b9a,color:#000
    classDef analysis fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef utility fill:#fff9c4,stroke:#f57f17,color:#000
    classDef external fill:#f5f5f5,stroke:#333,color:#000

    class Craft,Fix,Plan,Docs,Reverse primary
    class RepoSearch,Recon,Audit,Summary,Leaves,XV,DLC,Explorer,DocsSeeker evidence
    class Debug,Security,Scenario,ST,PS,Brainstorm,Predict analysis
    class Log,Org,Chrome,Cortex,Slide utility
    class User,Detect external

The intent table behind that graph:

Intent	Skill	Escalates to
Feature / implementation	`hi-craft`	`hi-plan` (no plan), `hi-fix` (3+ test failures)
Bug / failure	`hi-fix`	`hi-debug`, `hi-problem-solving`
Plan / architecture	`hi-plan`	`hi-repository-search --deep`, red-team, validate
Repository question / impact	`hi-repository-search`	evidence-only; the calling skill keeps orchestration
Debug investigation	`hi-debug`	`hi-codebase-research-explorer`, `hi-docs-seeker`
Security audit	`hi-security`	`mind_mcp` policy context, `graph_mcp` structure (T3)
Edge cases / tests	`hi-scenario`	`graph_mcp` feature paths (T3)
Ambiguous decision	`hi-brainstorm`	`hi-plan --two` for competing approaches
Risky change pre-check	`hi-predict`	verdict drives the next step: GO → `hi-plan`, CAUTION → mitigations, STOP → redesign
Documentation pack / wiki	`hi-maintenance-docs`, `hi-document-lifecycle`, `hi-repowiki`	discovery leaves, refinement, review resolution
C/C++ reverse engineering	`hi-reverse`	its own validation and package gates
Deck / pptx	`slide-kit`	—
Browser task	`hi-chrome-devtools`	—
Retrieval infrastructure	`hi-cortex`	`hi-repository-search` once MCPs are healthy

4. Agent Spawning And Delegation

Most DevKit skills stay single-agent by default. Delegation happens only when the skill file explicitly allows it and the work has independent tracks, repeated failures, or a mode that asks for parallel execution. Because DevKit runs on multiple harnesses (Claude Code, ZCode, Zed, VS Code Copilot) with different delegation primitives, all worker launches follow dev-shared/delegation-contract.md:

Probe once per workflow — check the runtime for a native subagent tool, then a delegation MCP server, then fall back to inline execution. Probe the tool list, not the harness name.
Roles are briefs, not agent types — researcher, implementer (alias fullstack-developer), reviewer, log-writer are briefs under dev-shared/roles/ (or skill references) pasted into whatever generic subagent the harness provides.
Fallback ladder — native spawn → MCP spawn → inline sequential passes, disclosed as [inline:<role>]; sequential self-analysis is never presented as parallel agents.
Receipts and guards — every delegation (or inline fallback) writes a receipt to the task primitive or log; max 5 delegated units per workflow; escalation depth 1.

Skill	Delegates?	Trigger / condition
`hi-codebase-research-explorer`	Yes, by design	Splits local, external, or hybrid exploration into `1-5` non-overlapping agents. Skips task registration when `<=2` agents or task tools are unavailable. Each agent has a 3-minute timeout.
`hi-craft`	Conditional	`--parallel` delegates one `implementer` (`fullstack-developer` alias) per phase. Test failures are handled directly for attempts 1-2; attempt 3+ invokes `hi-fix`. Planner and tester roles are never delegated.
`hi-fix`	Conditional	Standard/deep modes can activate `hi-codebase-research-explorer` or `2-3` parallel workers. `--parallel` creates a separate task tree and delegates `implementer` per independent issue.
`hi-plan`	Conditional	Delegates the `researcher` role: 1 for `--full`, 2 for `--hard` and `--parallel`, 2+ for `--two` (competing approaches). Fast/default mode never delegates.
`hi-debug`	Conditional	Multi-component investigations, parallel log/data collection, or CI/CD failures with 3+ possible causes may spawn parallel collection agents.
`hi-log`	Conditional	Delegates `log-writer` for material changes; writes inline for small sessions. If there is no material change, logging aborts.
`hi-repository-search`	Conditional	Does not spawn by default. Uses at most two investigators, code and documents, only when delegation is permitted and the work has independent tracks, spans 3+ subsystems, or needs independent conflict verification.
`hi-maintenance-docs`, `hi-document-lifecycle`, `hi-repowiki`	Conditional	Orchestrators may run independent domain leaves in parallel when their manifests do not overlap; the orchestrator keeps sole ownership of checkpoint and manifest state.

graph TD
    A["Task / workflow request"] --> B{"Delegation allowed by skill?"}
    B -->|no| C["Single-agent execution"]
    B -->|yes| D{"Independent tracks or explicit parallel mode?"}
    D -->|no| C
    D -->|yes| E["Spawn scoped subagents"]
    E --> F["Collect outputs<br/>dedupe / reconcile conflicts"]
    F --> G["Main agent owns synthesis<br/>integration and verification"]

    E --> H["hi-codebase-research-explorer<br/>1-5 scoped agents"]
    E --> I["hi-plan<br/>researchers by mode"]
    E --> J["hi-fix / hi-craft<br/>implementer or hi-fix escalation"]
    E --> K["hi-debug<br/>parallel evidence collection"]
    E --> L["hi-log<br/>log-writer"]
    E --> M["hi-repository-search<br/>max 2 investigators"]
    E --> N["orchestrators<br/>independent domain leaves"]

    classDef decision fill:#ffe0b2,stroke:#e65100,color:#000
    classDef primary fill:#c8e6c9,stroke:#1b5e20,color:#000
    classDef delegated fill:#bbdefb,stroke:#0d47a1,color:#000
    classDef synthesis fill:#e1bee7,stroke:#6a1b9a,color:#000
    class B,D decision
    class C,G primary
    class E,H,I,J,K,L,M,N delegated
    class F synthesis

5. HARD-GATEs

Skill	HARD-GATE	Violation behavior
`hi-craft`	Do not edit until a reviewed plan exists, unless the user explicitly skips planning.	Stop, create or reuse a plan, then continue.
`hi-fix`	Do not fix before locating the failure and identifying root cause.	Return to explore/diagnose; after 3 failed fix attempts, question architecture with the user.
`hi-plan`	Scan active plans first and record relevant `blockedBy` / `blocks` relationships.	Update the related plans before producing the new plan.
`hi-repository-search`	Provide traceable evidence only; do not invent context or own implementation decisions.	Report search coverage and ask for details if evidence is missing.
`hi-security`	Security findings need source evidence and severity-ranked remediation.	Do not auto-fix outside the requested audit/fix scope.
`hi-project-organization`	Do not override established project layout without evidence.	Use existing conventions or return an advisory recommendation.

Pipeline gates

Pipeline gates that behave like hard gates:

Skill	Gate
`hi-cross-artifact-validation`	Blocks unless at least two domain manifests are available; never modifies source artifacts.
`hi-maintenance-docs`	Publication requires an explicit approval artifact; resolving every comment does not approve a document.
`hi-document-lifecycle`	Pauses in `waiting_for_input`; resume revalidates input, artifact, and parent hashes. Retry and clarification counters never reset.
`hi-reverse`	`CATALOG_GATE` before planning, `REVERSE_PACKAGE_GATE` before a package is considered complete.

6. General Rules (Cross-cutting)

Hard-gate first, fast-path later — default to the lightweight mode, use flags for expansion.
Invoke > Delegate — invoke skills inline by default; delegate roles only on explicit triggers (3+ fails, 2+ hypothesis fails, large scope, --parallel, independent domain tracks). All delegation follows dev-shared/delegation-contract.md: probe once → role brief → receipt.
Token budget — each delegated role = 10-15K tokens. Prioritize inline methodology.
Test/Verify just enough — typecheck+lint for quick, +build+test for standard, comprehensive for deep, per dev-shared/test-budget.md.
Review optional — run only via --review or full mode. Auto-approve requires score ≥ 9.5 + 0 critical.
Finalize = commit + log — always conclude with a git commit + /hi-log (decisions, root causes, impacts).
Discovery artifacts are draft-only — leaves produce evidence-backed drafts; refinement works on isolated candidates; publication needs explicit human approval from the orchestrator.
Jev optional — typed judgment calls (routing, triage, scoring) may use dev-shared/jev/jev_ask.py per dev-shared/jev-contract.md. On probe failure or low confidence, fall through to the existing heuristic — behavior is identical with or without Jev.

Navigation Menu

Repository navigation

devkit.md

devkit.md

DevKit — Workflow Diagrams

0. Skill Roles

0.1 Orchestrators

0.2 Core workflow skills

0.3 Leaf skills (called by main skills)

0.4 Analysis and review skills

0.5 Standalone tool and integration skills

1. Core Pipelines

1.1 `hi-craft` — Feature Implementation

Mode Matrix

1.2 `hi-fix` — Issue Resolution

1.3 `hi-plan` — Implementation Planning

2. Discovery → Validation → Documentation

2.1 Discovery Artifact Pipeline

2.2 Documentation Lifecycle

2.3 Repository Wiki Pipeline

2.4 `hi-reverse` — self-contained reverse-engineering pipeline

3. Cross-skill Integration

4. Agent Spawning And Delegation

5. HARD-GATEs

Pipeline gates

6. General Rules (Cross-cutting)

FilesExpand file tree

devkit.md

Latest commit

History

devkit.md

File metadata and controls

DevKit — Workflow Diagrams

0. Skill Roles

0.1 Orchestrators

0.2 Core workflow skills

0.3 Leaf skills (called by main skills)

0.4 Analysis and review skills

0.5 Standalone tool and integration skills

1. Core Pipelines

1.1 hi-craft — Feature Implementation

Mode Matrix

1.2 hi-fix — Issue Resolution

1.3 hi-plan — Implementation Planning

2. Discovery → Validation → Documentation

2.1 Discovery Artifact Pipeline

2.2 Documentation Lifecycle

2.3 Repository Wiki Pipeline

2.4 hi-reverse — self-contained reverse-engineering pipeline

3. Cross-skill Integration

4. Agent Spawning And Delegation

5. HARD-GATEs

Pipeline gates

6. General Rules (Cross-cutting)

1.1 `hi-craft` — Feature Implementation

1.2 `hi-fix` — Issue Resolution

1.3 `hi-plan` — Implementation Planning

2.4 `hi-reverse` — self-contained reverse-engineering pipeline