fix: setup team persistence, interactive pager, blocked-by relations

[Finesssee]

Summary

• Terminal output is disabled from some commands, and not all teams are returned #34: Disable auto-pager for interactive commands (setup, auth, interactive, doctor) so keyboard input works and the terminal is not left raw on macOS; fetch all teams via cursor pagination in setup/interactive instead of a single unpaginated page.
• Configuring does not persist team settings #35: Persist default team from linear setup and linear config set default-team; document known config keys in help/show; use default team (or LINEAR_CLI_TEAM) when --team is omitted.
• relations add --relation blocked-by returns HTTP 400 (advertised but unsupported; should map to inverse 'blocks') #36: Map relations add --relation blocked-by to inverted Linear blocks (swap from/to) since Linear has no blockedBy enum value.

Test plan

• Unit tests for relation endpoint swap and blocked-by API mapping
• Unit tests for config keys / default_team TOML roundtrip
• Unit tests for pager disabled on interactive commands
• CLI help tests for config default-team and relations blocked-by
• Manual: linear setup on a large workspace — full team list, team saved, terminal still echoes after exit
• Manual: linear config set default-team ENG then i create without -t
• Manual: relations add A B -r blocked-by creates B blocks A (same as blocks with swapped args)

Closes #34
Closes #35
Closes #36

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

[Finesssee]

Thermo-nuclear review: PR #37

PR: #37
Scope: setup team persistence (#35), interactive pager + team pagination (#34), blocked-by relations (#36)
Verdict: FOLLOW_UP_REQUIRED — do not merge as-is

---

What's solid

• relations add --relation blocked-by returns HTTP 400 (advertised but unsupported; should map to inverse 'blocks') #36 root cause is correct. Linear has no blockedBy; invert-to-blocks is the right model. resolve_relation_endpoints is a good pure helper with tests.
• Terminal output is disabled from some commands, and not all teams are returned #34 pager diagnosis is correct. Auto-pager on interactive setup explains multi-enter + dead echo; opting out for setup/auth/interactive is the right fix.
• Configuring does not persist team settings #35 config surface finally matches the README. default_team on Workspace is the right place for per-profile defaults.
• Teams fetch now uses the canonical paginate_nodes path instead of inventing a second pager.

---

Blockers

1. src/config.rs crosses 1000 lines (724 → 1001)

This is exactly the skill's hard smell: a PR that pushes a file past 1k.

The new surface (get_default_team / set_default_team / key listing / tests) belongs in config, but the file was already a mixed bag of IO, OAuth, keyring, workspace CRUD, and CLI key dispatch. This PR made that worse by ~280 lines of behavior + tests.

Code judo: extract before merge, not after:

• src/config/mod.rs — load/save, Config / Workspace types
• src/config/team.rs (or defaults.rs) — default-team get/set + key table
• src/config/oauth.rs — already-adjacent OAuth helpers
• keep tests next to each module

Or at minimum: pull default-team + known-keys into a focused submodule so config.rs stays under 1k.

2. src/main.rs grew further (2314 → 2651) and still owns setup

handle_setup is a full product command living in the binary root, and this PR made it longer (pagination, selection-by-key, persistence messaging, JSON shape). main.rs was already a god file; adding more wizard logic is structural debt, not a fix.

Code judo: move setup into src/commands/setup.rs (same pattern as doctor, auth, interactive). Pager policy can stay as a thin fn command_disables_pager in main, or better: a method on Commands / trait so main doesn't grow a parallel command taxonomy.

ust // preferred shape impl Commands { fn disables_pager(&self) -> bool { matches!(self, Self::Setup | ...) } }

3. Workspace { ..., default_team: None } splat is spaghetti growth

Every construction site now hand-writes default_team: None. That is the textbook "special case bolted into every call site" pattern. Miss one later and you get a compile error today — good — but the design is still noisy and will keep growing as more workspace fields appear.

Code judo: give Workspace a constructor / Default-friendly shape:

ust impl Workspace { fn with_api_key(api_key: impl Into<String>) -> Self { Self { api_key: api_key.into(), oauth: None, default_team: None } } }

Then set_api_key / workspace_add / OAuth insert sites stop re-listing every field. Preserve-on-update should load the existing workspace and mutate fields, not rebuild with partial clones of oauth/team.

4. Default-team resolution is only half-wired (product/architecture leak)

get_default_team is used in:

• issues create
• interactive

…and nowhere else that already requires --team (sprint, statuses, cycles, metrics, import, templates create, triage, etc.). The user-facing promise of "I set a default team in setup" is false for most team-scoped commands.

That is not a nit — it is incomplete design. Either:

• A. Make default team a global fallback used by every optional/required team flag (canonical helper: resolve team = CLI arg → env → config), or
• B. Document that default team only applies to create/interactive and change the setup copy accordingly.

Right now we have A advertised, B implemented. That is a boundary leak between config and command layer.

---

High-conviction structural issues (not cosmetic)

5. Relation success printing re-implements invert logic

resolve_relation_endpoints already encodes the swap. The human message then re-branches on is_inverted() to pick response fields. That is duplicated model knowledge.

Simpler (behavior-preserving for the common case):

ust println!( "Created relation: {} {} {}", from, relation.display_name(), to );

If you want API-canonical identifiers (LIN-1 not raw UUID args), un-invert with the same helper instead of a second if-ladder:

ust let (issue_id, related_id, _) = resolve_relation_endpoints(from, relation, to); // then map response issue/relatedIssue back using that knowledge once

Prefer deleting the second invert branch entirely.

6. Teams-fetch pagination is copy-pasted twice

Setup and interactive both embed nearly identical GraphQL + paginate_nodes boilerplate. Canonical home is api (or a small teams helper used by both):

ust pub async fn fetch_all_teams(client: &LinearClient) -> Result<Vec<Value>>

Setup should call that; interactive should call that. One query string, one page-size policy.

7. Profile resolution is reimplemented again

get_default_team / set_default_team hand-roll:

ext LINEAR_CLI_PROFILE → config.current → "default"

…which already exists as current_profile() (Result) and is repeated across api-key paths. set_default_team can also create an empty workspace (api_key: "") if nothing is configured — half-applied state when the user hasn't authenticated yet.

Prefer: require an existing profile/workspace (error if none), reuse one profile resolver, mutate in place.

8. config get default-team vs config show disagree on semantics

• get → effective value (env wins via get_default_team)
• show → stored workspace value + separate env line

That dual model will confuse scripts and humans. Pick one:

• get = stored, document env as override only, or
• show/get both print effective with source annotation

---

Smaller maintainability notes (only if blockers fixed)

• command_disables_pager treating Doctor as interactive is debatable — doctor is mostly output, not password/TUI. Prefer pager-off only for true terminal owners (setup/auth/interactive/config set-key). Mild.
• Setup still prints output-format tips without persisting format — fine for Configuring does not persist team settings #35 scope, but the wizard still over-promises "Step 3" as configuration. Either persist or drop the step.
• to_api_string(BlockedBy) == "blocks" without swap is footgun if reused outside resolve_relation_endpoints. Consider making API type + endpoints one function only, and not exposing a misleading to_api_string for inverted variants.

---

What I would merge toward (minimal restructuring path)

1. Extract handle_setup → commands/setup.rs (main.rs stops growing).
2. Extract default-team + known keys so config.rs is back under 1k (module split).
3. Add Workspace::with_api_key / mutate-in-place updates; delete field-splat sites.
4. Add api::fetch_all_teams; delete duplicate queries.
5. Wire default team through a single resolve_team_arg(cli: Option<String>) -> Option<String> used by all team-flagged commands, or narrow the advertised contract.
6. Collapse relation display invert branch.
7. Keep the behavioral fixes for Terminal output is disabled from some commands, and not all teams are returned #34/Configuring does not persist team settings #35/relations add --relation blocked-by returns HTTP 400 (advertised but unsupported; should map to inverse 'blocks') #36 — they are right; the structure around them is not yet merge-grade.

---

Approval bar check

Criterion	Status
No clear structural regression	Fail — config.rs past 1k; main.rs larger god-file
No obvious missed simplification	Fail — Workspace splat, dual invert, dual team fetch
No unjustified file-size explosion	Fail — config 724→1001
No spaghetti special-case growth	Partial fail — default team only on 2 call sites
No wrong-layer / incomplete abstraction	Fail — default team not a real CLI-wide default
Behavior for the three issues	OK if tests + manual checks pass

Verdict: REQUEST CHANGES — fix the structural items above (especially 1–4), then re-review.