[feat] /goal persisted long-horizon workflows

[echobt]

Summary

Adds long-horizon persisted goals (/goal) to Cortex CLI / Cortex Code, stored with the session so an objective survives turn end, compaction, and TUI resume.

• /goal <objective> create/replace the active goal and kick off a plan→act→verify turn
• /goal status (objective, state, progress, budget)
• /goal pause / resume / clear (reserved tokens only when they are the entire argument)
• States: active | paused | complete | budget_limited | blocked
• Idle continuation while active and within budget (default 8 turns; wrap-up near the limit)
• Evidence-based completion via the UpdateGoal tool (reason + file/command/test evidence). The model cannot pause. Failed updates are atomic.
• goal.json next to session meta.json / history.jsonl
• Slash palette (after /plan) and composer chip in designer accent #1F4945

Lock v2 boards (designer + runtime) now include /goal PNG locks at 40×12 and 120×40:

• goal-chip-active / paused / done / budget / blocked — composer chip text-only (no radios)
• slash-palette with /goal after /plan

Live /goal smoke uses these env names only (never commit or log values):

Variable	Role
OPENAI_BASE_URL or CORTEX_LLM_BASE_URL	Astra /v1 (docs/test default http://84.32.63.4:20128/v1)
OPENAI_API_KEY or CORTEX_LLM_API_KEY	Operator-injected key
CORTEX_LLM_MODEL	cx/gpt-6-astra (gpt-astra 404s)

If no key is present, unit tests still pass and the live smoke prints SKIP: with that list. Agent mode still uses the Cortex API; this does not add a third product provider.

Test plan

• cargo fmt --all -- --check
• cargo clippy -p cortex-tui --all-targets -- -D warnings (pre-existing explicit_counter_loop in untouched files only)
• Targeted tests: lock_v2 + lock_v2_goal + lock_palette (goal chip copy, /goal after /plan at both sizes, unique frames, no retired palette)
• Lock PNGs unique at 36 (40×12) and 82 (120×40); accent #1F4945 present; 0 violet #A78BFA
• Full CI (./scripts/clippy.sh, cargo test --workspace, TUI job, audit, coverage)

Attestation (required)

I attest that:

• Security reviewed — auth, exec/sandbox, path traversal, network egress, and secret handling in this diff were reviewed. No secrets, tokens, or keyring dumps are in the change.
• Product-facing errors — user-visible failures use Cortex product copy. API-down paths say The coding service is temporarily unavailable. No raw provider, SDK, or transport names.
• TUI verified — every TUI surface touched in this PR was exercised (build + headless snapshot / ratatui-style test, and a real run when a TTY was available).
• Tests added — unit tests cover the new logic; TUI changes include a snapshot or buffer assertion. No mocks that report success.
• No secrets — no API keys, WorkOS secrets, R2/AWS credentials, or .env files are included.

Risk

Protocol adds Op::Goal and EventMsg::GoalUpdated. Session goal.json is local metadata only. Live smoke is skip-by-default without a key. Compatible /v1 env names are aligned; Cortex Agent auth is unchanged. Quality line-count debt on lock_v2.rs is reduced by extracting scene ids.

  

[greptile-apps]

Too many files changed for review (102 files, 100 file limit).

Bypass the limit by tagging @greptile-apps to review.