orchestration-envelope
Skill23 starsChanged 50 days agoMITView on GitHub
Internal machine-result contract for headless agentic-workflow drivers. The executable source is @gtrabanco/agentic-workflow-schema: strict Envelope v2 for workflow-status, compact SkillOutcome v1 for driven work, compatibility parsing, and deterministic document snapshots. Not a menu entry.
In this repository30 other files it gives its agents.
- AGENTS.mdAGENTS.md
- template/AGENTS.mdAGENTS.md
- audit-docsSkill
- audit-prSkill
- batch-schedulerSkill
- discover-repository-stateSkill
- execute-phaseSkill
- fold-findingsSkill
- init-workspaceSkill
- log-sessionSkill
- phase-contractSkill
- pre-execution-reviewSkill
- product-auditSkill
- resolve-repository-stateSkill
- review-a11ySkill
- review-brandSkill
- review-changeSkill
- review-codeSkill
- review-debtSkill
- review-designSkill
- review-implementationSkill
- review-perfSkill
- review-securitySkill
- review-seoSkill
- review-verifySkill
- skill-lintSkill
- triage-issueSkill
- unit-laneSkill
- verification-contractSkill
- workflow-statusSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 52
Read the profile and append renderOutputInstruction(skill) only for a driven invocation.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Machine result contract (internal)
Interactive skills remain text-first. A headless driver obtains a compact, validated result at the boundary; it does not add a repeated JSON section to every user-facing skill.
The canonical Turn contract remains here for the executor and review skills that load it.
The executable source of truth is
@gtrabanco/agentic-workflow-schema:
types, JSON Schemas, renderOutputInstruction(skill), parseTurn(input), and
compileWorkflowSnapshot(input). This document states the policy only; do not
copy a second schema here.
Output profiles
WORKFLOW_SKILL_PROFILES is the authoritative inventory.
workflow-statusalways returns the strict Envelope v2 sensor result. Its envelope includes the detailed project view underdetail.- The other driver-invoked skills return SkillOutcome v1 only when the
driver appends
renderOutputInstruction(skill)to that invocation. It has the small model-owned fields: outcome, next intent/targets, blockers, questions, discoveries, and evidence references. advance(the companion pi package command) is the conductor, not a worker profile: it runs as deterministic code over the sensor envelope and never emits a worker SkillOutcome.- Interactive invocations emit their normal human-readable reports; no driver result is required.
Both results are one final fenced json block. The package rejects unknown
keys at the routing boundary. Repository facts are never reconstructed from
model prose: the driver compiles WorkflowSnapshot v1 from its selected,
versioned documents and caller-supplied repository facts.
Driver protocol
- Read the profile and append
renderOutputInstruction(skill)only for a driven invocation. - Pass the final response to
parseTurn({skill, text, context}). Keep the returned source and diagnostics with the run journal. - On an absent, malformed, or invalid machine result, re-invoke the same
session once with:
Emit only the machine result for the turn above. - Parse the repair reply. A second failure is driver-level
FAILED; never retry indefinitely and never turn arbitrary prose into workflow facts.
Compatibility is deliberately narrow: it can repair documented legacy v2 shapes only when the missing value is mechanically knowable. A nonzero issue count without issue identities, an unmatched numeric unit id, or unrelated prose remains invalid and is surfaced to the driver.
Contract evolution
- Envelope v2 is strict for new drivers.
detailis required (usuallynull) and skill-specific extensions live inside it;design_candidates, for example, isdetail.design_candidates, never a root key. workflow-statusretains Envelope v2 for existing sensor consumers.parseEnvelope()remains the legacy-compatible package API; new consumers useparseEnvelopeV2Strict()orparseTurn().SkillOutcome v1andWorkflowSnapshot v1are separate, versioned JSON Schemas. A breaking change to any published contract is a package major.
Normalized Repository State
Drivers call discover-repository-state before planning and retain the frozen
docs/workflow/REPOSITORY_STATE.md reference. WorkflowSnapshot v1 preserves
unknowns, provenance, and declared contradictions; a driver routes a
contradiction to resolve-repository-state rather than silently replacing it.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

