fix-remove-openrouter — Workgraph live mirror

Metadata

Status	done
Assigned	`agent-1338`
Agent identity	`f51439356729d112a6c404803d88015d5b44832c6c584c62b96732b63c2b0c7e`
Created	2026-05-01T14:51:51.936346967+00:00
Started	2026-05-01T14:53:42.132019428+00:00
Completed	2026-05-01T15:39:46.138466412+00:00
Tags	`fix,config,defaults,cleanup`, `eval-scheduled`
Eval score	0.91
└ blocking impact	0.93
└ completeness	0.95
└ constraint fidelity	0.10
└ coordination overhead	0.92
└ correctness	0.90
└ downstream usability	0.90
└ efficiency	0.85
└ intent fidelity	0.90
└ style adherence	0.92

Description

A fresh wg init (claude-cli, codex-cli, local, nex-custom routes) writes an [openrouter] section into the config. The dispatcher's registry-refresh job then probes OpenRouter every ~30s; without an API key it errors continuously. User reports 25+ consecutive failures filling the daemon log with no operational impact — pure log spam.

User question 2026-05-01: 'should we remove it as a default part of the config?'

Answer: yes. [openrouter] should only be in the config when the user explicitly chose the openrouter route OR added it via wg endpoint add openrouter .... For users on claude / codex / nex routes, OpenRouter has no purpose.

Scope

Routes affected

wg init --route claude-cli — should NOT write [openrouter]
wg init --route codex-cli — should NOT write [openrouter]
wg init --route local — should NOT write [openrouter]
wg init --route nex-custom — should NOT write [openrouter]
wg init --route openrouter — should write [openrouter] (this is the openrouter route, of course)

Verify each route's dry-run output and remove the [openrouter] section from the routes that aren't openrouter.

Existing user configs

Per the user's 'skip back-compat ceremony' standing rule (memory: feedback_skip_backcompat_ceremony.md):

For the user's local config: remove [openrouter] directly as part of this task
wg migrate config should also recognize and offer to remove the section if no openrouter usage is detected (idempotent — running it on a config that already has it removed is a no-op)

Registry refresh side

Even when [openrouter] IS present (because the user wants it), the registry-refresh failure mode should be quieter:

After N consecutive failures (e.g., 5), log ONCE and stop trying for ~1h (cooldown)
On manual config reload OR API key set, retry immediately

This is a small backoff/circuit-breaker for the noisy log; matches the design-exponential-failure design (filed earlier today) for failure cycles.

Validation

Failing test: wg init --route claude-cli --dry-run output does NOT contain [openrouter]
Same for codex-cli, local, nex-custom routes
wg init --route openrouter --dry-run DOES contain [openrouter] (don't break the path that needs it)
User's local config /home/erik/workgraph/.wg/config.toml has [openrouter] section removed (this task does the surgery directly)
wg migrate config recognizes orphaned [openrouter] sections (no api key, no usage in profiles or endpoints) and offers to remove. Idempotent.
Registry refresh failure surfacing: cooldown after N failures (don't spam the log every poll)
cargo build + cargo test pass
cargo install --path . was run before claiming done

## Description
A fresh `wg init` (claude-cli, codex-cli, local, nex-custom routes) writes an `[openrouter]` section into the config. The dispatcher's registry-refresh job then probes OpenRouter every ~30s; without an API key it errors continuously. User reports 25+ consecutive failures filling the daemon log with no operational impact — pure log spam.

User question 2026-05-01: 'should we remove it as a default part of the config?'

Answer: yes. `[openrouter]` should only be in the config when the user explicitly chose the openrouter route OR added it via `wg endpoint add openrouter ...`. For users on claude / codex / nex routes, OpenRouter has no purpose.

## Scope

### Routes affected
- `wg init --route claude-cli` — should NOT write [openrouter]
- `wg init --route codex-cli` — should NOT write [openrouter]
- `wg init --route local` — should NOT write [openrouter]
- `wg init --route nex-custom` — should NOT write [openrouter]
- `wg init --route openrouter` — should write [openrouter] (this is the openrouter route, of course)

Verify each route's dry-run output and remove the `[openrouter]` section from the routes that aren't openrouter.

### Existing user configs
Per the user's 'skip back-compat ceremony' standing rule (memory: feedback_skip_backcompat_ceremony.md):
- For the user's local config: remove `[openrouter]` directly as part of this task
- `wg migrate config` should also recognize and offer to remove the section if no openrouter usage is detected (idempotent — running it on a config that already has it removed is a no-op)

### Registry refresh side
Even when [openrouter] IS present (because the user wants it), the registry-refresh failure mode should be quieter:
- After N consecutive failures (e.g., 5), log ONCE and stop trying for ~1h (cooldown)
- On manual config reload OR API key set, retry immediately

This is a small backoff/circuit-breaker for the noisy log; matches the design-exponential-failure design (filed earlier today) for failure cycles.

## Validation
- [ ] Failing test: `wg init --route claude-cli --dry-run` output does NOT contain `[openrouter]`
- [ ] Same for codex-cli, local, nex-custom routes
- [ ] `wg init --route openrouter --dry-run` DOES contain `[openrouter]` (don't break the path that needs it)
- [ ] User's local config `/home/erik/workgraph/.wg/config.toml` has `[openrouter]` section removed (this task does the surgery directly)
- [ ] `wg migrate config` recognizes orphaned `[openrouter]` sections (no api key, no usage in profiles or endpoints) and offers to remove. Idempotent.
- [ ] Registry refresh failure surfacing: cooldown after N failures (don't spam the log every poll)
- [ ] cargo build + cargo test pass
- [ ] cargo install --path . was run before claiming done

Depends on

done .assign-fix-remove-openrouter

Required by

(none)

Log

2026-05-01T14:51:51.861073890+00:00 Task paused
2026-05-01T14:52:26.435706820+00:00 Task published
2026-05-01T14:53:37.171922781+00:00 Lightweight assignment: agent=Careful Programmer (f5143935), exec_mode=full, context_scope=task, reason=Careful Programmer matches implementation + testing requirements; Careful tradeoff suits config changes requiring exhaustive multi-route verification and idempotent migration logic.
2026-05-01T14:53:42.132024137+00:00 Spawned by coordinator --executor claude --model opus
2026-05-01T14:53:57.159889+00:00 Starting work — exploring config init code paths
2026-05-01T15:38:42.841252557+00:00 Validated: dry-run check across all 5 routes — only openrouter route emits [openrouter]
2026-05-01T15:38:49.798925219+00:00 Validated: wg migrate config detects + removes orphaned [openrouter], preserves it when in use, idempotent
2026-05-01T15:38:49.832492531+00:00 Added registry-refresh circuit breaker (5 failures → 1h cooldown). Logs first error + threshold trip event only — no more 25+ identical errors
2026-05-01T15:38:49.868849452+00:00 Tests: 5 new route presence/absence tests + 3 migrate idempotency tests + 2 circuit-breaker tests, all passing
2026-05-01T15:38:49.907334830+00:00 cargo build + cargo test --bin wg pass (3308 tests). 1 pre-existing lib failure (test_global_config_path) unrelated to this change
2026-05-01T15:38:49.966105244+00:00 cargo install --path . done — verified installed binary's dry-run output is correct
2026-05-01T15:39:37.571663657+00:00 Committed: 68a098f93 — pushed to remote
2026-05-01T15:39:46.138475649+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
2026-05-01T15:43:27.357673480+00:00 PendingEval → Done (evaluator passed; downstream unblocks)