SDLC Orchestrator Architecture

Prompt-first procedure: Describe the outcome you want in your agent conversation. The agent should select and load the appropriate AIWG assets, explain material changes, request any needed approval, and report verification evidence. Exact commands and flags appear only in the CLI reference.

Purpose

AIWG maps natural-language SDLC requests to canonical skills, then coordinates the work with the active provider's native capabilities. The architecture is provider-neutral: no deployed command directory or provider-specific agent tool defines workflow behavior.

Authority Model

User intent
  -> sdlc-quickref (always-loaded routing)
  -> the agent capability search "<capability>"
  -> the agent stable-asset loader skill <name>
  -> canonical SKILL.md
  -> referenced playbook/templates/rules
  -> provider-native execution
  -> .aiwg/ artifacts and gate evidence

Authority descends from the canonical skill into explicitly referenced executable playbooks, templates, and rules. Generated slash commands are adapters or compatibility shims. A deployed command may invoke a workflow but does not supersede its skill.

Discovery and Selection

1. Apply `sdlc-right-sizing` so small work does not accidentally expand into a full lifecycle flow. 2. Consult the kernel-resident `sdlc-quickref` for common capability domains. 3. When the exact skill is not already identified, run the agent’s capability search. 4. Surface the selected result and fetch it with the agent’s stable-asset loader. 5. Read the complete skill and required referenced resources before execution.

Do not enumerate the workflow catalog from memory and do not inspect `.claude/commands/flow-*.md` as the primary source.

Natural-Language Mapping

User intent	Discovery phrase
Transition to Elaboration	`SDLC transition to Elaboration`
Run a security review	`SDLC security review cycle`
Create an architecture baseline	`SDLC architecture baseline`
Run an iteration	`SDLC iteration dual track`
Assess current project progress	`project status`

Explicit adapter input such as `/flow-inception-to-elaboration` maps to the same-named canonical skill and follows that skill's current contract.

Execution Model

The selected skill declares inputs, outputs, gates, constraints, and references. Some skills reference a machine-readable playbook that carries executable sequence and gate semantics. Providers should honor that contract using their available primitives.

For collaborative artifact work, a skill may use this logical pattern:

Primary Author -> Independent Reviewers -> Synthesizer -> Baseline/Archive

The mechanics vary by provider. Claude may expose a `Task`-style API, Codex may expose collaboration primitives, and other providers may execute or emulate the pattern differently. Steward/provider guidance determines the mechanism. Independent work can run concurrently within configured caps; dependent work remains sequential.

Artifacts and Gates

• Stage temporary work under the locations declared by the skill, commonly `.aiwg/working/<workflow>/`.
• Write final artifacts to their appropriate `.aiwg/` category.
• Maintain requirements-to-code-to-test-to-deployment traceability.
• Preserve human-authorization requirements and phase gates.
• Report progress, decisions, blockers, verification evidence, and final artifact paths.

Adapter Contract

Provider adapters must:

1. Resolve to the corresponding canonical skill. 2. Preserve declared parameters such as the guidance option or the interactive option only when the skill supports them. 3. Avoid embedding divergent workflow behavior. 4. Be regenerated or corrected when they conflict with the canonical skill.

Adapters may be retained for convenience and backward compatibility. They are not the authoring surface for new workflow behavior.

Maintenance Checks

When changing orchestration behavior:

• Update the canonical skill or referenced playbook first.
• Keep `sdlc-quickref`, discovery metadata, and rules aligned.
• Regenerate assembled project context and provider adapters.
• Verify the agent’s capability search and the agent’s stable-asset loader expose the intended contract.
• Search active assets for command-first or provider-specific guidance that contradicts the authority model.

References

• `@$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/skills/sdlc-quickref/SKILL.md`
• `@$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/skills/orchestrate-project/SKILL.md`
• `@$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/rules/sdlc-orchestration.md`
• `@$AIWG_ROOT/agentic/code/addons/aiwg-utils/rules/skill-discovery.md`
• `@$AIWG_ROOT/agentic/code/addons/aiwg-utils/skills/steward/SKILL.md`