Problem
ADR-014 ("knownFlags-snapshot mechanism") recorded an architectural decision about how the manifest stores feature flags. That mechanism was superseded by the FlagsRecord key-presence approach in PR #300 (feat/flags-typed-registry).
The decisions ledger still describes the retired mechanism, which misleads future developers consulting ADR-014.
Current Status
Implementation Constraint
This is a Learning-pipeline task, not a Code-agent edit:
decisions.md is a rendered artifact generated from .devflow/learning/decisions-ledger.jsonl
- Direct edits to
decisions.md are overwritten on the next Learning agent run
- Only the Learning agent can amend or retire the ADR via the
retire-anchor or assign-anchor ledger operations (per LLM-vs-plumbing principle)
Solution
Route this work through the Learning agent:
- Add an observation to the learning queue indicating that ADR-014 is stale
- The Learning agent should either:
- Retire ADR-014 (if the old mechanism is fully obsolete and no longer applicable)
- Amend ADR-014 (add a note explaining how FlagsRecord supersedes the old mechanism, and describe the new approach)
- Re-render decisions.md via the Learning agent's
render-decisions operation
Acceptance Criteria
Related
Problem
ADR-014 ("knownFlags-snapshot mechanism") recorded an architectural decision about how the manifest stores feature flags. That mechanism was superseded by the FlagsRecord key-presence approach in PR #300 (feat/flags-typed-registry).
The decisions ledger still describes the retired mechanism, which misleads future developers consulting ADR-014.
Current Status
knownFlags-snapshotfield in manifestFlagsRecordkey-presence (adopted in PR feat(flags): typed flag registry, settings-page TUI, new upstream flags #300)Implementation Constraint
This is a Learning-pipeline task, not a Code-agent edit:
decisions.mdis a rendered artifact generated from.devflow/learning/decisions-ledger.jsonldecisions.mdare overwritten on the next Learning agent runretire-anchororassign-anchorledger operations (per LLM-vs-plumbing principle)Solution
Route this work through the Learning agent:
render-decisionsoperationAcceptance Criteria
Related
.devflow/learning/decisions-ledger.jsonl(source of truth).devflow/learning/decisions.md(rendered output)