D0
Skill stevennitesh/programming-agent-skills/docs/validation/evals/to-tickets-prompt4/inputs/d0
Practical workflow skills for coding agents: scope, debugging, cleanup, subagents, GitHub tracking, and verification.
npx -y skills add stevennitesh/programming-agent-skills --skill d0Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Create and publish approved, dependency-ordered ready-for-agent tickets from settled source material.
SKILL.md
6.8 KB, as published. Nobody here has run it
To Tickets
Own one outcome: one approved, dependency-ordered graph of independently grabbable ready-for-agent tickets from one bounded body of settled source, with exhaustive coverage and one verified next action.
docs/agents/issue-tracker.md and its pointers own tracker transport, the
Ready-for-agent contract, roles, relationships, ready queries, packet
mechanics, and Mutation read-back. The engineering contract owns shared Source
Trace, proof-seam, and state-boundary rules. This skill owns admission,
coverage, slicing, approval, publication scope, and Return.
Admit only settled implementation source with an identifiable owner and
publication authority, stable outcome, decisions, scope, exclusions, and proof
expectations. Apply the setup gate through the target repo's AGENTS.md. Return
the exact $repo-bootstrap precondition when its tracker surface is absent or
incompatible.
Process
1. Trace
Trace the complete bounded source: the request and approvals, repo instructions, every supplied parent artifact, decision-bearing comments, and directly required durable owners. Record source identity and owner, outcome, accepted decisions, deferrals and rejections, scope and exclusions, evidence, proof expectations, and material gaps.
Return one complete source-gap packet and stop before slicing when a missing or conflicting decision could change intent, contracts, architecture, supported state, migration, acceptance, proof, or scope. Leave implementation-owned technique to implementation.
2. Map
Inspect repository reality only far enough to name stable seams, supported state branches, proof lanes, expected production scope, overlap, migration constraints, and durable domain, ADR, glossary, or prototype pointers.
Leave patch design, exact file selection, helpers, and test implementation to the delivery owner. Return a source conflict rather than letting current code silently override settled intent.
3. Slice
Local proof policy: every ticket has observable proof through its claimed boundary, and completion can be judged without unrelated sibling completion. This independence is a To Tickets completion gate; it does not establish concurrency.
Every ticket satisfies the tracker's Ready-for-agent contract and adds:
- parent or bounded-source reference;
- work-unit form and migration phase when applicable;
- why this slice, what to build, and covered commitments; and
- relevant Source Trace and durable context pointers.
Acceptance names observable behavior and the highest meaningful proof seam.
Each ticket also names true blockers or none, expected write scope, semantic
ownership, shared resources and serial tripwires, and a scope fence. Split only
when distinct proof, authority, rollback, dependency unlock, permission, state,
migration, or ownership makes separate completion valuable; do not split by
layer, filename, team, or available agent slot.
For stateful work, include the engineering contract's state-boundary matrix: put every applicable branch and any evidenced non-applicable axis in acceptance and its proof lane.
Apply the coverage gate: map every source-visible implementation commitment and scope boundary to a ticket, an explicit deferral or exclusion, or a no-ticket reason.
If exhaustive coverage requires no ticket, return the coverage map and no-ticket result without approval or mutation. Otherwise Slice completes only when every commitment has exactly one disposition, every ticket has source justification, and the graph, predicted frontier, and applicable execution profiles are coherent.
4. Approve
Present one identified proposal revision containing the Source Trace summary, complete coverage map, ordered graph, predicted frontier, intended tracker mutations, and every ticket's form and roles, blockers and consumed outcome, commitments, acceptance, proof, rationale, write scope, parallel safety, applicable execution profile and state matrix, and scope fence.
Obtain explicit approval of that exact revision before publication. A material change to coverage, a ticket, edge, acceptance, proof, state, scope, readiness, or relationship requires reconciliation and fresh approval. Return the complete proposal awaiting approval when approval is absent.
5. Publish
Immediately before mutation, reconcile the source, proposal, tracker target, and parent relationships. Material drift returns to Approve. Publish only the fresh approved parent, tickets, relationships, roles, state, and packet metadata, blockers first, through the tracker owner.
Preserve the parent's intent and lifecycle. Change its body only for approved, tracker-required child or ordering metadata. Stop before implementation, review, claim, or closeout.
Apply Mutation read-back to the parent, ordered children, bodies, roles, state, relationships, blocking edges, affected dependents, and resulting ready frontier. Provider receipts do not prove completion. A failed, unknown, or mismatched mutation is blocked; return the approved revision, observed applied and failed operations, unknown state, frontier risk, and safest non-duplicating recovery.
For a completely verified graph, return its references, packet path when applicable, coverage summary, ready frontier, and exactly one next action in this priority order:
- empty frontier: resolve the named blocker;
- explicitly requested non-empty parent-delivery run: recommend
$parallel-implementwith the parent, regardless of initial frontier width; - one ready ticket: recommend
$implementfor it; - overlapping semantic ownership, production scope, public seam, fixture,
proof resource, or serial tripwire: recommend
$implementfor the first ticket in tracker order; - at least two substantial, semantically independent, production-isolated,
proof-isolated tickets: recommend
$parallel-implementwith the parent; or - uncertain independence or economics: recommend
$implementfor the first ticket in tracker order.
Recommend and stop; never begin the selected implementation route.
Return
Return exactly one typed result: setup precondition, source-gap packet, no-ticket result, proposal awaiting approval, partial-publication recovery, or published graph. Each result names its evidence, unchanged or observed tracker state, and exact safe continuation.
Completion
Successful publication requires setup, source, coverage, proposal, approval, freshness, publication, and read-back gates; tracker-owned readiness and this skill's artifact fields; distinct blockers and serial constraints; applicable state matrices and execution profiles; an exact observed graph without partial failure; and one verified next action returned without starting it. Earlier typed Returns are bounded stops, not successful publication.