chelch5/ticket-pack-builder

Ticket Pack Builder

Use this skill to create or refine the repo-local work queue. Use ../../references/competence-contract.md as the backlog-quality bar: a ticket is not good enough if its acceptance leaves the operator trapped between scope boundaries and a literal failing closeout command.

Modes

• bootstrap: generate the first implementation-ready backlog during the initial scaffold
• refine: regenerate, expand, or normalize an existing backlog later
• remediation-follow-up: add or normalize canonical follow-up tickets after audit, diagnosis, review, or repair work identifies concrete remediation or reverification needs

Mode selection

• If this skill is reached from scaffold-kickoff during greenfield scaffolding, use bootstrap.
• If this skill is reached from retrofit or later backlog maintenance work, use refine.
• If audit, diagnosis, generated review, or repair outputs already identify specific remediation or reverification work, use remediation-follow-up.
• If invoked directly and tickets/manifest.json does not exist yet, use bootstrap.
• If invoked directly and a backlog already exists, use refine unless the request is explicitly about post-audit or post-repair follow-up, in which case use remediation-follow-up.
• If the user request is ambiguous about whether to create the first backlog or revise an existing one, ask the user before choosing a mode.

Bootstrap mode procedure

1. Read the finalized generation surfaces

Read:

• docs/spec/CANONICAL-BRIEF.md
• .opencode/skills/
• .opencode/agents/
• .opencode/commands/
• .opencode/state/workflow-state.json

Identify:

• What features/capabilities need to be built
• What infrastructure/setup is required
• What declared target platforms require dedicated export or release-proof lanes
• What the acceptance criteria are
• Which areas are blocked on unresolved decisions
• The backlog readiness signal
• The finalized repo-local validation commands and workflow surfaces

2. Break work into implementation waves

Organize tickets into waves based on dependency order:

• Wave 0: Foundation — repo setup, environment bootstrap, dependency installation, configuration
• Wave 1: Core — the primary functionality the project exists to provide
• Wave 2: Secondary — supporting features, integrations, secondary workflows
• Wave 3: Polish — error handling hardening, performance, documentation, UX refinement

Core-wave discipline:

• If the product is a client for an external API, LLM provider, local inference runtime, database, browser, or engine service, create explicit core tickets for that connectivity before release proof exists.
• Do not let a parser/router/shell ticket masquerade as core product delivery when the real backend path is still unwritten.
• A user-facing command, tool gateway, provider adapter, IDE surface, agent runtime, or other product-spine ticket must own real runnable behavior for its declared scope, not only type shells or command registration.
• If a broad ticket title would otherwise need stubbed behavior, split the backlog into a narrower scaffold-only ticket plus the real behavior ticket instead of normalizing stubs inside the primary ticket.

3. Create individual tickets

For each piece of work, create a ticket with these fields:

• id — unique identifier (e.g., SETUP-001, CORE-001, FEAT-001)
• title — short descriptive title
• wave — which execution wave this belongs to
• lane — which project area or ownership lane this belongs to
• parallel_safe — whether this ticket can be advanced in parallel with other tickets when dependencies are satisfied
• overlap_risk — low, medium, or high expected overlap with other tickets
• stage — planning (all new tickets start here)
• status — todo or blocked for new tickets; later lifecycle-aligned queue values are derived by ticket_update
• resolution_state — open for new tickets, later done, reopened, or superseded
• verification_state — trusted only after current proof exists; initialize new tickets as suspect
• depends_on — list of ticket IDs this depends on
• source_ticket_id — source ticket when this is a follow-up or remediation ticket, otherwise null
• follow_up_ticket_ids — linked downstream remediation or expansion tickets, initially empty
• source_mode — process_verification, post_completion_issue, net_new_scope, or split_scope when the ticket was created from later diagnosis, reverification work, or open-parent decomposition; omit for greenfield bootstrap tickets
• split_kind — required when source_mode is split_scope; use sequential_dependent when the child must not run until the parent-owned work is complete, or parallel_independent when the child can truly advance independently; default to sequential_dependent unless project evidence proves the child has no dependency on the parent's open work
• For EXEC-REMED-001 follow-up tickets that repair missing runnable review evidence on an open parent ticket, treat the child as parallel_independent: the parent stays open, but the evidence-repair child is the foreground work and must not deadlock behind the parent it is meant to unblock
• finding_source — original audit, review, QA, or smoke finding code when this ticket exists to remediate a validated issue
• summary — one-paragraph description of what needs to be done
• acceptance — list of specific acceptance criteria tied to finalized repo-local commands, checks, or observable workflow surfaces; these must be scope-isolated to the ticket's own work
• artifacts — empty list (populated during execution)
• decision_blockers — list of unresolved decisions that block this ticket (empty if none)

Ticket sizing rules

• Each ticket should be completable in ONE agent session
• If a ticket requires changes to more than 5-7 files, split it
• If a ticket has more than 5 acceptance criteria, consider splitting
• Prefer many small tickets over few large ones

Parallel lane rules

• Mark parallel_safe: true only when the ticket has no unresolved dependency on another unfinished ticket and the expected overlap risk is low
• Use overlap_risk to make concurrency judgment explicit instead of assuming it from lane names
• Default to parallel_safe: false when the ownership boundary is unclear
• Keep lane names aligned with real write ownership so lease-based execution has a credible boundary

External phase grouping

• Keep tickets/manifest.json canonical at ticket granularity even when an adjacent orchestration service later batches several tickets into one downstream phase.
• Keep orchestration-owned phase grouping, PR numbers, reviewer assignment, and merge-policy state out of tickets/manifest.json and .opencode/state/workflow-state.json.
• Do not create fake merge, PR, or review tickets just to mirror GitHub workflow. Those are orchestration overlays, not repo-local queue truth.
• Make ticket IDs and acceptance criteria specific enough that a downstream PR can always point back to the exact owning ticket set and proof requirements.

Mandatory Wave 0 bootstrap ticket

Wave 0 must always include one bootstrap/setup ticket that covers:

• runtime and toolchain presence
• project dependency installation
• test dependency installation
• verification command readiness
• initial canonical brief and scaffold sanity checks

Treat that ticket as the default first active ticket unless the canonical brief proves a different setup sequence is required.

Mandatory target-completion tickets for declared Tier 1 release targets

If the canonical brief or bootstrap provenance declares a Tier 1 release target, create dedicated tickets for the target-completion path instead of burying that work inside generic polish.

For Godot Android repos, always create:

• ANDROID-001 in lane android-export
• SIGNING-001 in lane signing-prerequisites (only when the brief requires packaged delivery; see Product Finish Contract below)
• RELEASE-001 in lane release-readiness

ANDROID-001 acceptance must cover:

• export_presets.cfg exists with an Android preset
• repo-local android/ support surfaces exist and are non-placeholder
• the canonical Android export command is recorded in the ticket and repo-local skills

SIGNING-001 acceptance must cover (when packaged delivery is required):

• keystore ownership decision is made by the project team and recorded in the brief
• keystore path, alias, and password are available to the build environment
• signing mode (debug-only vs release-signed) is explicitly resolved in canonical truth

RELEASE-001 acceptance must cover:

• godot --headless --path . --export-debug "Android Debug" build/android/<project-slug>-debug.apk or the exact resolved Godot-binary equivalent succeeds (runnable proof)
• the APK exists at build/android/<project-slug>-debug.apk
• unzip -l confirms Android manifest plus compiled classes or resources content
• when the finish contract requires packaged delivery: a signed release APK or AAB exists and SIGNING-001 is closed

RELEASE-001 must declare depends_on edges to two kinds of prerequisites:

1. Infrastructure gate: SIGNING-001 when the brief requires packaged delivery. In debug-only mode, omit SIGNING-001.
2. Feature gate (required): all terminal product tickets — the highest-wave tickets that are not in infrastructure lanes (android-export, signing-prerequisites, release-readiness) and not in process lanes (remediation, reverification). When the backlog has no such tickets, note the gap as a decision blocker on RELEASE-001.

source_ticket_id records split-scope lineage for provenance only. It is not a depends_on target. The workflow engine forbids a split-scope child from blocking on its source_ticket_id (deadlock rule). RELEASE-001's source_ticket_id is ANDROID-001 or SIGNING-001; neither must appear in depends_on.

Do not let a generic POLISH-001, UX, or validation ticket stand in for Android export or release proof ownership.

Mandatory vertical-slice tickets for backend-backed products

If the canonical brief describes a product whose user-facing surfaces depend on a real backend or provider path, seed explicit tickets for that integration before the release lane.

Examples include:

• LLM or external API clients
• local inference or model runtimes
• IDE or agent shells that must talk to a live control plane
• tool gateways whose value depends on real execution instead of only schema registration

For these repos:

• create at least one ticket that proves a real user-facing entrypoint reaches the actual backend/runtime path
• keep that ticket in Wave 1 or Wave 2, not in an optional post-release bucket
• make release-readiness depend on those vertical-slice tickets
• keep acceptance criteria runnable and concrete, such as a real CLI invocation, provider-backed integration check, or end-to-end request/response proof

Do not author tickets that say or imply:

• "without yet filling every command body with final product behavior"
• "keep command bodies shallow"
• "justified stubs"
• "deferred to later tickets" for the core runtime behavior of the same ticket's declared surface

If the real behavior truly belongs later, rename the current ticket to the narrower scaffold-only scope and create the later behavior ticket immediately in the backlog.

Mandatory finish-ownership tickets for consumer-facing repos

If the canonical brief includes a Product Finish Contract (section 13) that forbids placeholder output in the final product, create explicit finish-ownership tickets instead of burying finish work in a generic polish bucket.

Finish work must be split into owned tickets that cover:

• finish direction or style decision if unresolved (create as a blocked or decision ticket)
• visual content production or integration (one ticket per distinct visual domain when scope warrants separation)
• audio content production or integration when the brief requires audio deliverables
• final finish validation against the recorded contract

Do not create one generic FINISH-001 or POLISH-001 ticket for all content work. The backlog must be specific enough for an agent to determine what constitutes completion for each owned area.

When the finish contract explicitly allows placeholder output in the shipped product (for example placeholder_policy: placeholder_ok), do not invent finish-ownership tickets for those content areas.

Leaving finish work as unwritten commentary outside the canonical backlog is not permitted when the Product Finish Contract records a non-placeholder finish bar.

Handling unresolved decisions

• Do NOT fabricate implementation detail for work that depends on unresolved major choices
• Create explicit blocked tickets with decision_blockers listing what needs to be decided
• Create decision tickets for choices the team needs to make (e.g., "Choose database engine")
• Create discovery tickets for research that needs to happen before implementation

Spec coverage check

Before writing any ticket files, verify that the ticket backlog covers every feature area in the canonical brief:

1. List every feature, capability, or required output from the canonical brief's Goals and Required Outputs sections
2. For each feature area, confirm that at least one ticket exists whose acceptance criteria address that feature
3. Any brief feature with no ticket coverage must either:
   • (a) become its own new ticket with spec-derived acceptance criteria, or
   • (b) be explicitly listed as a blocked decision ticket with a named follow-on that will resolve it
4. Record the coverage check result as a short table in the canonical brief appendix or as a comment in the manifest: feature → ticket_id

This check prevents the greenfield run from emitting a backlog that silently omits whole spec areas. A backlog that reaches this step with uncovered spec features must add tickets — it cannot proceed to manifest writing until coverage is complete.

4. Write ticket files

For each ticket, write a markdown file to tickets/<id>.md using the template in tickets/templates/TICKET.template.md.

5. Update the manifest

Write tickets/manifest.json with the structure defined in references/ticket-system.md:

• version: 3
• project: project name from canonical brief
• active_ticket: first ticket in wave 0
• tickets: array of all ticket objects
• keep ticket objects aligned with the runtime contract in .opencode/lib/workflow.ts

6. Generate the board

Write tickets/BOARD.md as a human-readable view of the backlog organized by wave, showing:

• Ticket ID, title, status, dependencies
• Grouped by wave
• This is a DERIVED VIEW — the manifest is the source of truth

7. Align workflow state

Ensure .opencode/state/workflow-state.json reflects the first active ticket:

{
  "active_ticket": "<first-ticket-id>",
  "stage": "planning",
  "status": "todo",
  "approved_plan": false,
  "ticket_state": {
    "<first-ticket-id>": {
      "approved_plan": false,
      "reopen_count": 0,
      "needs_reverification": false
    }
  },
  "process_version": 7,
  "process_last_changed_at": null,
  "process_last_change_summary": null,
  "pending_process_verification": false,
  "parallel_mode": "sequential",
  "bootstrap": {
    "status": "missing",
    "last_verified_at": null,
    "environment_fingerprint": null,
    "proof_artifact": null
  },
  "lane_leases": [],
  "state_revision": 0
}

Output contract

Before leaving this skill, confirm all of these are true:

• tickets/manifest.json exists, uses version: 3, and every ticket record matches the runtime ticket contract
• tickets/BOARD.md exists and is clearly derived from the manifest instead of carrying extra machine state
• every tickets/<id>.md file exists for the manifest entries created or changed in this run
• .opencode/state/workflow-state.json names the foreground ticket from the manifest and seeds bootstrap.status: "missing" on fresh scaffolds until bootstrap proof exists
• any declared Tier 1 target platform has its canonical export or release-proof lane present in the manifest before handoff
• follow-up tickets preserve source_ticket_id, follow_up_ticket_ids, and source_mode linkage when the work came from diagnosis, repair, or reverification evidence
• remediation tickets preserve finding_source so downstream review, QA, and closeout can rerun the original failing check
• every feature area in the canonical brief's Goals and Required Outputs has at least one ticket with matching acceptance criteria (spec coverage check completed)

Refine mode

Use when expanding or normalizing an existing backlog:

1. Read the existing manifest
2. Identify gaps, unclear tickets, or new work from updated brief
3. Add/modify tickets following the same rules
4. Regenerate BOARD.md

Remediation-follow-up mode

Use when audit, diagnosis, review, or repair work already produced concrete findings and the repo needs canonical follow-up tickets.

Public follow-up command when a diagnosis pack already contains ticket recommendations:

python3 skills/ticket-pack-builder/scripts/apply_remediation_follow_up.py <repo-root> --diagnosis <diagnosis-dir-or-manifest>

1. Read the source evidence first

Collect the current source of truth for the follow-up:

• the latest diagnosis pack in diagnosis/<timestamp>/ when working from scafforge-audit
• the current repair summary and recorded process-version state when working from scafforge-repair
• the latest registered review, qa, smoke-test, or backlog-verification artifacts when working inside a generated repo
• any machine-readable recommended-ticket payload emitted alongside the audit or diagnosis reports

Do not infer remediation work from vague commentary when canonical evidence already exists.

2. Decide whether to create, reopen, or defer

For each finding:

• create a new follow-up ticket when the work is net-new and scoped
• reopen or reverification-route an existing ticket when the historical ticket already owns the defect and current evidence justifies restoring or revising trust
• defer ticket creation when the evidence is incomplete or intent-changing and requires a human decision first

3. Preserve canonical linkage

For every remediation or reverification ticket:

• set source_ticket_id when the work came from a prior ticket
• append the new ticket ID to the source ticket's follow_up_ticket_ids
• set finding_source when the follow-up exists to remediate a validated audit, review, QA, or smoke-test finding
• keep verification_state: suspect until current evidence proves the issue resolved
• record the evidence source in the ticket summary or Notes so downstream agents can trace the follow-up back to the diagnosis, review, or repair artifact
• ensure the ticket acceptance reruns the original finding-producing command when the repo already has a concrete build, lint, reference, or smoke failure to clear

4. Respect guarded ticket creation

When the generated repo exposes guarded ticket tooling:

• route creation through the canonical ticket_create flow instead of raw manifest edits when the workflow contract requires it
• preserve any source-mode or provenance fields that distinguish process verification, post-completion defects, or net-new scoped work

When operating at package level without the generated tooling available, emit the canonical ticket payload and update the generated ticket surfaces in the same contract shape the repo expects.

5. Regenerate derived views

After applying the follow-up:

• update tickets/manifest.json
• regenerate tickets/BOARD.md
• ensure any changed trust, reverification, or process-follow-up state stays consistent with .opencode/state/workflow-state.json

6. Emit canonical repair completion evidence when this ran as repair follow-up

When scafforge-repair required ticket-pack-builder as a repair follow-on stage, read .opencode/meta/repair-follow-on-state.json and, after the ticket follow-up work is actually complete, write:

• .opencode/state/artifacts/history/repair/ticket-pack-builder-completion.md

Use this minimal shape so the public repair runner can auto-recognize completion for the current repair cycle on the next run:

# Repair Follow-On Completion

- completed_stage: ticket-pack-builder
- cycle_id: <cycle_id from .opencode/meta/repair-follow-on-state.json>
- completed_by: ticket-pack-builder

## Summary

- Created or updated the canonical repair follow-up tickets required by the current repair cycle.

Do not write this artifact speculatively. Only emit it once the remediation-follow-up ticket work is actually complete for the current cycle.

After this step

Continue to ../handoff-brief/SKILL.md as directed by scaffold-kickoff.

Rules

• Keep manifest machine-readable
• Keep board human-readable (derived from manifest)
• Keep queue status coarse and aligned with the workflow tools: todo, ready, plan_review, in_progress, blocked, review, qa, smoke_test, done
• Do NOT use ticket status for transient approval state (that's in workflow-state.json)
• Treat plan_review and smoke_test as workflow-tool-owned queue values, not as free-form authoring choices
• Keep closeout acceptance criteria aligned with the deterministic smoke_test tool rather than generic PASS prose
• When a ticket needs a smoke gate, prefer one explicit backticked repo-local smoke command so smoke_test can treat that command as the canonical smoke scope instead of improvising a heuristic subset or full-suite fallback
• Keep acceptance criteria scope-isolated. If a literal closeout command would still fail on later-ticket work, split the backlog differently or encode the dependency explicitly instead of shipping contradictory acceptance
• Keep wave, lane, parallel_safe, and overlap_risk aligned with real execution boundaries
• Record dependencies explicitly
• Put acceptance criteria on every ticket
• Prefer executable acceptance criteria where possible so downstream agents have concrete repo-local commands or observable pass/fail checks to run
• Do not generate a ticket whose literal acceptance command knowingly reaches into sibling-ticket or later-ticket scope
• Do not generate a product-spine ticket whose summary, acceptance, or Notes normalize shallow shells, placeholder responses, justified stubs, or deferred runtime behavior
• Keep historical completion separate from current trust: status stays queue-oriented, while resolution_state and verification_state represent historical closure and present trust
• Treat post-audit and post-repair ticket creation as a first-class workflow path, not an ad hoc backlog note
• Keep ticket docs, workflow docs, and ticket tools aligned on the same lifecycle semantics before handoff
• Do not store orchestration-owned phase or PR metadata in the manifest or workflow-state

References

• references/ticket-system.md for the contract
• assets/templates/TICKET.template.md for the package-side canonical ticket template; it must stay identical to the generated tickets/templates/TICKET.template.md surface