impl-agency-csv-roundtrip — Workgraph live mirror

Metadata

Status	done
Assigned	`agent-2390`
Agent identity	`f51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e`
Model	`codex:gpt-5.5`
Created	2026-05-04T17:41:12.121183869+00:00
Started	2026-05-04T18:46:37.222795738+00:00
Completed	2026-05-04T19:12:52.233188040+00:00
Tags	`agency,sync,impl,federation`, `eval-scheduled`
Eval score	0.88
└ blocking impact	0.93
└ completeness	0.86
└ constraint fidelity	0.70
└ coordination overhead	0.91
└ correctness	0.90
└ downstream usability	0.91
└ efficiency	0.82
└ intent fidelity	0.74
└ style adherence	0.86

Description

Implement byte-for-byte round-trippable CSV import/export matching agentbureau/agency's 12-column starter schema (type, name, description, quality, domain_specificity, domain, origin_instance_id, parent_content_hash, scope, parent_ids, generation, created_by). Driven by findings from research-agency-csv-roundtrip. Depends on hash-compat (so re-hash logic must already be settled).

Likely scope

Extend wg agency import to accept a --format agency-csv flag and parse all 12 columns (CSV → primitive YAML).
Add wg agency export --format agency-csv to emit a CSV byte-equal to agency's starter.csv when given a primitive set.
Treat parent_ids as JSON array column (per spec). Treat domain as comma-separated.
Add fixture tests/fixtures/agency-starter-sample.csv (5–10 rows from real agency starter).
Round-trip test: parse → serialize → byte-diff against fixture.

File scope

src/commands/agency_import.rs
src/commands/agency_push.rs (for export)
src/agency/store.rs (only if needed for canonical serialization helpers)
tests/fixtures/agency-starter-sample.csv
tests/integration_agency_csv_roundtrip.rs

Do NOT touch:

src/agency/hash.rs (owned by impl-agency-hash-compat)
src/agency/types.rs (owned by impl-agency-schema-fields)

Validation

Failing test written first: test_agency_csv_byte_equal_roundtrip
All 12 agency columns round-trip cleanly
Domain comma-separated and parent_ids JSON-list parsing covered
cargo build + cargo test pass with no regressions
Live smoke: wg agency import --format agency-csv tests/fixtures/agency-starter-sample.csv then wg agency export --format agency-csv - byte-equals the input

## Description
Implement byte-for-byte round-trippable CSV import/export matching agentbureau/agency's 12-column starter schema (type, name, description, quality, domain_specificity, domain, origin_instance_id, parent_content_hash, scope, parent_ids, generation, created_by). Driven by findings from `research-agency-csv-roundtrip`. Depends on hash-compat (so re-hash logic must already be settled).

## Likely scope
- Extend `wg agency import` to accept a `--format agency-csv` flag and parse all 12 columns (CSV → primitive YAML).
- Add `wg agency export --format agency-csv` to emit a CSV byte-equal to agency's starter.csv when given a primitive set.
- Treat parent_ids as JSON array column (per spec). Treat domain as comma-separated.
- Add fixture `tests/fixtures/agency-starter-sample.csv` (5–10 rows from real agency starter).
- Round-trip test: parse → serialize → byte-diff against fixture.

## File scope
- src/commands/agency_import.rs
- src/commands/agency_push.rs (for export)
- src/agency/store.rs (only if needed for canonical serialization helpers)
- tests/fixtures/agency-starter-sample.csv
- tests/integration_agency_csv_roundtrip.rs

Do NOT touch:
- src/agency/hash.rs (owned by impl-agency-hash-compat)
- src/agency/types.rs (owned by impl-agency-schema-fields)

## Validation
- [ ] Failing test written first: test_agency_csv_byte_equal_roundtrip
- [ ] All 12 agency columns round-trip cleanly
- [ ] Domain comma-separated and parent_ids JSON-list parsing covered
- [ ] cargo build + cargo test pass with no regressions
- [ ] Live smoke: `wg agency import --format agency-csv tests/fixtures/agency-starter-sample.csv` then `wg agency export --format agency-csv -` byte-equals the input

Depends on

done .assign-impl-agency-csv-roundtrip

Required by

(none)

Log

2026-05-04T17:41:12.062515439+00:00 Task paused
2026-05-04T18:37:42.133653599+00:00 Task published
2026-05-04T18:38:09.861150293+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=This is a correctness-sensitive implementation with file edits and roundtrip tests, so the Careful Programmer is the best fit over the evaluator/assigner roles.
2026-05-04T18:46:37.222800948+00:00 Spawned by coordinator --executor codex --model gpt-5.5
2026-05-04T18:46:51.274893437+00:00 Starting implementation; no unread messages and no prior branch commits found
2026-05-04T19:06:46.446158701+00:00 Validated: cargo build passes; cargo test --test integration_agency_csv_roundtrip passes; live agency_csv_roundtrip smoke passes; full cargo test has unrelated pre-existing provider/config failures noted
2026-05-04T19:06:46.568883166+00:00 Implemented 12-column agency-csv import/export, byte-equal fixture roundtrip test, and smoke scenario
2026-05-04T19:12:05.036152705+00:00 Committed: 5252a6d17 — pushed to remote
2026-05-04T19:12:52.233201577+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
2026-05-04T19:14:03.871653281+00:00 PendingEval → Done (evaluator passed; downstream unblocks)