workflow-chaining

workflow-chaining - Generic Declarative Workflow Runner

Role

Read the YAML, decide the workflow plan once, announce each phase, dispatch specialists via the Skill tool (NL description-match as fallback), write checkpoints. Invoked by odoo-intake, a workflow command, or a plan's workflow node.

Hard rules

1. Prefer the Skill tool for phase dispatch. NL description-match is the fallback when no matching skill name is known at plan time.
2. NEVER spawn a sub-agent directly (no agent launch, no context: fork) except fan-out (≤3 workers, mandatory worker-brief preamble) and an agent: phase: launch that agent at the phase's model_tier; a later phase resumes it by the returned id.
3. No recursion. Fork workers are leaf agents and carry: "Do NOT invoke spawner skills via the Skill tool. Do NOT spawn sub-agents. You MAY use the Skill tool for read-only leaf skills (e.g. odoo-feature-check, odoo-override-finding). Only Read/Grep/Glob/Write/Bash."
4. One plan decision, then no stops. No phase runs before the workflow plan is decided (Phase 0 step 6). Phases then run without stopping; a phase gate is a text confirmation before that phase's outward or irreversible action, and the only one.
5. Resume from checkpoint. If resume: true and <output_dir>/<slug>-state.json exists (output_dir resolves per Phase 0 step 2), load it and skip done phases.
6. Schema → workflows/_schema.md.
7. on_complete EMITs, never dispatches. Matched transitions go to Continuation Contract next[] for the plan runner (the main agent) - this skill never fires a spawner itself.

Phase 0 - Load and validate

1. Read the referenced workflows/<name>.workflow.yaml file.
2. Resolve the run's state root once, via the resolve-capture-substitute protocol in ${CLAUDE_PLUGIN_ROOT}/snippets/state-root-resolution.md: output_dir: from the YAML is a relative .odoo-ai/<name> literal that this skill resolves against the Tier-2 ISOLATE dir (call mcp__plugin_odoo-ai-agents_odoo-local__project_dir with axis isolate, cwd = the run's target root) at RUNTIME - never write to the bare project-relative .odoo-ai/<name> path. Resolve series + profile per ${CLAUDE_PLUGIN_ROOT}/snippets/project-facts-resolution.md.
3. Orphan sweep (do this every invocation, before the resume check below). Run find <resolved ISOLATE output_dir>/ -mindepth 1 -maxdepth 1 -type d -mmin +43200 -exec rm -rf {} + (it removes whatever output_dir THIS workflow declares that was untouched for over 30 days). Full rule + bound: ${CLAUDE_PLUGIN_ROOT}/snippets/visual-evidence-lifecycle-contract.md § Clause 3. Run it unconditionally, every run.
4. If resume: true, check for <output_dir>/<slug>-state.json under the resolved ISOLATE dir; if found, load it and determine the last completed phase.
5. Collect any missing inputs[] from the user (one question per missing input).
6. Decide the workflow plan (${CLAUDE_PLUGIN_ROOT}/snippets/planning-gate-contract.md § One Plan-Mode gate per request):
   • A phase with plan: true computes the plan: run the phases up to and including it first (they write only under the state root), and present its output as the plan.
   • The workflow owns the request (invoked by odoo-intake, a command, or the user): call EnterPlanMode, present the block below, ExitPlanMode on approve; refine: [feedback] re-plans, cancel stops.
   • Every on_complete entry of the workflow hands the request to odoo-solution-design or odoo-planning (the workflow only scopes): that skill holds the request's Plan-Mode gate. State the chain in one line and run.
   • A node of an approved plan: present nothing and run.