Work Item Sequencing

Work Item Sequencing

Numbering convention for order + parallelism among work items in one epic. Number = prefix in Subject__c, space-delimited. Per-epic scope; cross-epic deps go in WI body prose, not numbers.

Companion to gus-cli (CLI mechanics: query, create, update, statuses, IDs).

Convention

Prefix Subject__c with a dotted number + single space:

1 Set up the runtime service
2 Migrate the parser to the runtime
1.1 Add config schema
1.2 Add config loader

• Sequential (1, 2, 3): in order. 2 waits on 1; 3 waits on 2.
• Parallel (1.1, 1.2): same-parent siblings, no waiting on each other.
• Group gating: next number at a level waits for all work under prior numbers. 2 waits on 1 + every 1.x/1.x.y.
• Arbitrary depth (1.1.1, …): same recursive rule.

The rule

WI ready when every WI sorting before it (at shallower-or-equal ancestor level) is done. Siblings parallel; sibling-group gates next number at parent level.

1.2 ready once 1.1 + descendants done. 2 ready once everything under 1 done.

Done

Done = Status__c is Closed or Completed.

Fixed is not done — usually means PR not merged yet. Non-terminal statuses (Fixed, Ready for Review, QA In Progress) never satisfy a prerequisite. See gus-cli ## Status__c values.

When to number

Optional.

• Use: epic-level planning, or when user states a dependency.
• Skip: standalone work; don't renumber an epic of accumulated independent items.

Unnumbered = no deps, nothing depends on it. Always ready; never gates, never gated. Don't infer order from date/position — no number means independent.

ADRs sequence first

Epic needing a new or updated ADR (per ADR-FORMAT "When to offer": hard-to-reverse + surprising + real trade-off) → the ADR-writing WI(s) get the lowest number: 1 (single ADR) or 1.1/1.2 (parallel siblings). Dependent WIs start at 2. Rationale: dependents encode a decision; the ADR must exist first so downstream work builds on a recorded decision, not an unstated one.

Before creating the rest of the epic, confirm the required ADRs with the user — which decisions need recording, repo-wide (docs/adr/) vs package-local (packages/*/docs/adr/). Then sequence the ADR WI(s) as 1/1.x and everything else 2+.

No ADR needed → number normally.

Assigning (planning)

1. Lay out work items.
2. Must-precede → sequential top-level (1, 2).
3. Parallel under a step → same parent, distinct suffix (1.1, 1.2).
4. Independent → leave unnumbered.
5. Prefix Subject__c with number + space. Create via gus-cli.

Dense from 1 per group preferred; gaps harmless (sort is by value).

Reading ("what's ready/unblocked in this epic?")

1. Query epic WIs incl Status__c. Query without open-only filter — done prerequisites are Closed, which open-only excludes (see gus-cli "What's unfinished in this epic").
2. Parse leading dotted number from each Subject__c. None → unnumbered.
3. Readiness: unnumbered → ready; numbered → ready iff all prior-sorting ancestors done.
4. Report 3 groups: Ready now (unblocked numbered + all unnumbered), Blocked (name blocking number), Done (optional).

Sort by integer segments (1.2 before 1.10, not lexical). Unnumbered sort last.

Edge cases

• Malformed (1., 1..2, 1.x): treat unnumbered + flag — likely typo'd dependency.
• Duplicate (two 1.2): allowed = parallel siblings. Don't flag.
• Orphan child (1.1, no 1): rule still holds; don't require parent to exist.
• Non-sequence number (W-12345 backport, version 2.40 release): only leading N/N.N… + space counts. Ambiguous → ask.