fix(quota): bound CLI todo diagnostics

[huangruiteng]

Summary

• compact repeated agent todo diagnostic lanes in the default quota should-run CLI projection after the full decision is computed
• preserve decision-relevant counts and bounded identities: selected/executable work, up to three unclaimed candidates, due or capability-blocked monitors, one monitor schedule-gap repair target, and up to three succession-warning todo ids
• expose the full diagnostic projection through --include-todo-summary-detail
• tighten quota output budgets and add scale plus Markdown/JSON semantic coverage

Qualification

• representative live loopx-meta readback at head fc915448: full detail 102,868 bytes; compact default 68,048 bytes (33.9% reduction)
• default JSON now retains unclaimed_open_count plus bounded unclaimed_priority_open_items; Markdown renders their todo ids
• default and detail modes preserve exact interaction_contract, scheduler_hint, and selected_todo
• focused pytest: 12 passed
• loopx canary premerge --from-git-diff: passed, 17/17 selected checks, no failures or manual holds
• public/private boundary scan: clean; no raw state, logs, credentials, or local paths committed

This is a bounded hot-path reduction. Repeated whole-backlog, current-agent duplicate, other-agent diagnostic, handoff, deferred, and completed-item lists remain on the explicit detail path; their decision-relevant counts stay hot.

[huangruiteng]

Resolved the succession-warning identity gap found in review.

Changed surfaces

• The compact quota should-run projection now keeps up to three public-safe
  todo_ids for todo_succession_warning while full warning items remain on
  the detail path.
• Markdown renders those bounded identities and retains the existing item-list
  fallback for unprojected payloads.
• A semantic regression test proves count=4 renders concrete repair todo IDs
  instead of todo_ids=n/a.

Validation

• Focused pytest: 10 passed.
• Ruff: passed for all changed files.
• Python compile: passed for all changed files.
• loopx canary premerge --from-git-diff: 17 selected, 17 passed, 0 failures,
  0 warnings, 0 manual holds.
• Public/private boundary scan and git diff --check: passed.

This coverage is sufficient because the change is limited to the CLI-only
projection and Markdown renderer, while the canary set exercises the broader
quota, scheduler, monitor, output-budget, and control-plane contracts.

[huangruiteng]

Updated after semantic review of the compact hot path.

Changed surfaces:

• retained up to three unclaimed_priority_open_items in default JSON and rendered their todo ids in Markdown; the previous compact output exposed only unclaimed_open_count
• retained one monitor_schedule_gap_items repair target and rendered its identity
• preserved the earlier succession-warning identity repair
• merged current main non-destructively; PR is mergeable at fc915448

Readback:

• full detail: 102,868 bytes
• compact default: 68,048 bytes
• actual compact output shows the two current unclaimed todos with status, task class, priority, and action kind

Validation:

• focused pytest: 12 passed
• risk-based premerge: 17/17 selected checks passed, zero failures, zero manual holds
• diff hygiene, changed Python compile checks, and public/private boundary scan passed

Coverage is sufficient for this PR because it proves both budget behavior and the identities needed for selection/repair while leaving repeated diagnostic lanes on the explicit detail path.

[huangruiteng]

Follow-up qualification: default Markdown identity parity

markdown is the CLI default (loopx/cli.py), while heartbeat and host integrations explicitly request JSON. JSON remains the machine-readable control-plane contract, but default Markdown is still a supported agent/operator interface and must remain actionable.

The current head therefore preserves bounded identity in both formats without duplicating the hot path:

• selected work reuses the existing agent_lane_next_action identity and falls back to selected_todo only when that lane is absent;
• agent_todo_next now reads the bounded executable lane and includes todo_id;
• the first three unclaimed todos retain todo_id, status, priority, task class, and action kind;
• due monitor, capability-blocked monitor, schedule-gap repair, and successor-warning identities remain bounded;
• counts and interaction_contract remain visible.

The first draft added duplicate selected/executable lines and the base/head output differential rejected the growth. The revised renderer reuses existing lines instead of raising the budget.

Validation on head 59779472:

• 12 passed: quota CLI projection and CLI output budget tests;
• Ruff passed on the renderer and focused tests;
• cli-output-budget-regression-smoke.py passed;
• loopx canary premerge --from-git-diff passed: 17/17 selected checks, no failures, skips, warnings, or manual holds;
• public/private scan and git diff --check passed.

This covers the changed quota projection, Markdown renderer, output-budget contract, and default/cold-path parity without expanding the output allowance.

[huangruiteng]

Residual projection qualification

Exact-head measurement against the live public-safe loopx-meta state showed that the compact agent_todo_summary was still 10,032 compact JSON characters.

The remaining duplication was concrete:

• first_open_items retained three monitor objects (about 3 KB), while the first item was also present in the dedicated monitor_due_items lane;
• every retained item repeated text as title and carried non-decision diagnostics such as generic update metadata.

Head 060228cc now:

• removes generic first_open_items from the default projection and records it in the omitted-lane detail metadata;
• keeps the bounded first_executable_items, unclaimed, due monitor, capability-blocked monitor, and schedule-gap lanes;
• projects retained items through an explicit action/routing/monitor field allowlist;
• leaves the full objects available through quota should-run --include-todo-summary-detail;
• tightens the focused compact-payload ratchet from 10 KB to 7 KB.

Readback against the same state:

• agent_todo_summary: 10,032 -> 6,093 compact characters;
• full default payload: 55,931 -> 51,992 compact characters;
• selected todo, executable candidate, unclaimed candidates, due monitor, counts, succession IDs, and interaction semantics remain present.

Validation:

• focused pytest: 12 passed;
• Ruff and git diff --check: passed;
• CLI base/head output-budget differential: passed;
• risk-based premerge canary: 17/17 passed, with no failures, skips, warnings, or manual holds.