tasks: held work, kept branches and spend limits say what is true

[santoshkumarradha]

Re-cut from #1604 (only the task-run and spend-limit fixes; the other #1604 pieces are elsewhere or already on dev).

• session: interactive chat bypasses daily spending limit #1546 (Part of): a chat turn or a chat task past today's spending limit now asks first with a card, "Raise to $X" or "Stop for today", instead of a bare refusal or a task that says "started" and then sits held. A raise lasts for today only and the top bar shows it at once. Dev already refused a chat turn past the limit; the card and the task-admission ask are what was missing.
• session: tasks stuck in working on restart, stopped status mapped to failed, and reader handoff #1554 (Part of): a helper started by the read hand-off no longer leaves a stopped card when the hand-off fails. Item 1 (a restart reading a never-run task as done) stays open. The other session: tasks stuck in working on restart, stopped status mapped to failed, and reader handoff #1554 items were already on dev.
• Manager brief taken by a reading helper that fakes the team's work #1568 (Fixes): the read hand-off helper is now read-only, so it cannot write, commit or merge, and a request it cannot serve goes back to the conversation.
• 'Run this as a task on <model>' becomes a reading helper; chat misreports it #1569 (Part of): the read-only bound is in; the chat's own prose about model and cost is unchanged.
• Checker spend ceiling is shared by the whole run, and the run still says done #1572 (Fixes): each check has its own spend ceiling, and a run whose checks did not finish no longer ends as done; the landing names the unfinished checks.
• Run ends 'branch kept' with nothing blocking; /land says nothing is waiting #1574 (Fixes): a kept task or run branch is listed by /land with the reason it was kept, /land reaches the engine over the local host, and editor .orig backups never reach a task branch.
• A standing firing that changed nothing is logged as landed #1582 (Fixes): a scheduled firing that only reported a sentence comes to "said", not "landed".

Not included: #1527, #1549, #1566, #1579 (already on dev), and the part-check path refusal (#1573, taken back in #1604).

Tested revision: b3b5d7d. CI (check, light gate, touched packages, CLA) passed on this head.

BASE=origin/dev make pr-ready on Spark fails on one test, TestAPreExistingRedDoesNotStandBetweenTheWorkAndItsLanding in the sharded session suite. It passes when run alone, and plain origin/dev fails it too under make test PKGS=./internal/session, along with two other timing tests, so it looks load-related. Test changes made to match the new behavior: three standing tests now expect "said", and dev's old refuse-the-turn daily-limit test is removed because the turn now waits on a raise-or-stop card. I did not show every new test failing without its fix.

[santoshkumarradha]

PR bar audit — directive #1790715358822565124

1. CI + changelog: ✅ all checks green (light gate, touched packages, check, license/cla). ✅ one unreleased changelog entry present (1670-task-run-audit.md).

2. Review passes: ❌ Pass 1 — ARCHITECTURE: absent. ❌ Pass 2 — SECURITY: absent. ❌ Pass 3 — CODE EFFICIENCY & DESIGN: absent.

3. Video proof: ❌ no video/GIF embedded in the PR body. This change is user-facing (held work, kept branches and spend limits say what is true) — the bar requires a capture of the change actually running, embedded in the body. To be captured on Spark.

4. Merge state: CONFLICTING against dev — needs a rebase; CI must be re-run green on the rebased head.

Fixes for the gaps are in flight. This PR does not turn ready until all four conditions hold.

_{drafted with CodeAF}

[santoshkumarradha]

PR-bar audit: CI green + changelog present — those hold. Missing:

1. Review passes — architecture / security / efficiency & design, none posted yet.
2. On-screen proof — held work, kept branches and spend limits are person-facing task surfaces; the bar requires the Spark tmux capture embedded in the body.

Marked draft until both close.
_{drafted with CodeAF}

[santoshkumarradha]

Bar audit — gaps against the new PR bar

1. CI + changelog: ⚠️ green but stale: run 36505663619 passed (check, light gate, touched packages, license/cla) on 2026-09-29; dev has moved since, so a fresh green gate against current dev is required before ready. ✅ changelog entry present (1670-task-run-audit.md).

2. Review passes: ❌ none of the three are posted (Pass 1 ARCHITECTURE, Pass 2 SECURITY, Pass 3 CODE EFFICIENCY & DESIGN), each as a separate comment naming what was checked and what was found.

3. Video proof: ❌ absent. User-facing change (task page: held work, retry/done wording, landcmd rendering). Needs a Spark tmux capture rendered to GIF/video, embedded in this PR body.

4. Spark: all verification ahead of ready runs on the Spark (fleet).

Action taken: converted back to draft until fresh CI, the three passes, and the body video exist.

—
Drafted with CodeAF · reviewed and owned by the author

[santoshkumarradha]

PR Bar Audit — gaps against the new bar

1. CI + changelog: ✅ CI fully green (check, light gate, touched packages, license/cla). ✅ Changelog present (docs/changes/unreleased/1670-task-run-audit.md).
2. Review passes: ❌ None posted — Pass 1 (ARCHITECTURE), Pass 2 (SECURITY), Pass 3 (CODE EFFICIENCY & DESIGN) all absent.
3. Video proof: ❌ Missing — user-facing change (held work / kept branches / spend-limit status text on screen); no video or GIF in the body.
4. Spark runs: captures/tests must run on the Spark (fleet).
_{drafted with CodeAF}

[santoshkumarradha]

Audit against the new PR bar

• CI + changelog: all checks green (check, light gate, touched packages, license/cla) ✅ — changelog entry present ✅
• Review passes: 0 of 3 — ARCHITECTURE, SECURITY, and CODE EFFICIENCY & DESIGN, each as its own comment naming what was checked and what was found.
• Video proof: missing — the spend-limit card ("Raise to $X" / "Stop for today") and the held-work/kept-branch messages are user-facing; a capture must be embedded in the PR body.

Converted back to draft until all four conditions hold; nothing merges until then.

_{drafted with CodeAF}

[santoshkumarradha]

PR-bar audit (team bar, 2026-10-03) - status per category:

• CI: 1 of 4 checks green, 3 still running, none failing.
• Changelog: present (1 entry).
• Review passes: 0 of 3 - the three labeled pass comments are absent; they are being posted (66-file diff).
• Video proof: required (user-facing TUI: held-work/land-command visibility) and MISSING - returning to draft per the no-video rule.

Stays draft until all four conditions hold; nothing merged. Fixes incoming: three review passes, then Spark capture.

_{drafted with CodeAF}

[santoshkumarradha]

PR bar audit — gaps against the new bar

1. CI + changelog — ✅ all checks green (light gate, touched packages, check, license/cla). ✅ changelog entry present (docs/changes/unreleased/1670-task-run-audit.md).

2. Review passes — all three missing (Pass 1 ARCHITECTURE, Pass 2 SECURITY, Pass 3 CODE EFFICIENCY & DESIGN), as separate comments naming what was checked and what was found.

3. Video proof — missing. User-facing change (held work, kept branches and spend limits must say what is true). A Spark tmux capture of the corrected status behavior, rendered to GIF/video, must be embedded in the PR body.

4. Runs on Spark — per the fleet skill.

Converted back to draft until all four conditions hold. Nothing merges before that.

_{drafted with CodeAF}

[santoshkumarradha]

Audit against the PR bar — what is missing

1. CI + changelog — ✅ every check green (check, light gate, touched packages, license/cla); ✅ changelog entry present.
2. Three review passes — ❌ none posted yet. Pass 1 (ARCHITECTURE), Pass 2 (SECURITY) and Pass 3 (CODE EFFICIENCY & DESIGN) follow as three separate comments, each naming what was checked and what was found.
3. On-screen proof — ❌ user-facing change (task rows: held work, kept branches, spend limits): no capture in the body yet.
4. Fleet runs — builds, tests and captures run over ssh on the fleet, never a laptop. Spark is offline at audit time, so the dumb fleet host stands in until it is back.

Also missing: labels and milestone (existing sets only).
Disposition — converted from ready back to draft until all four conditions hold; nothing merges.

_{drafted with CodeAF}

[santoshkumarradha]

Audit against the new PR bar:

• CI: green on its last run, but the branch now conflicts with current dev — rebase and re-run before trusting it.
• Changelog: present.
• Review passes: none posted. Owed: Pass 1 — Architecture, Pass 2 — Security, Pass 3 — Code Efficiency & Design.
• Video: none in the body; the task rows and spend-limit surfaces are user-facing, so a capture is owed.

Stays draft.

_{drafted with CodeAF}

[santoshkumarradha]

Bar audit — what is missing (per the new PR bar)

Audited at head b3b5d7d9c16ded140280ace1d341dcb935c3e7a2 by @tags.

1. CI + changelog — light gate ✅ · touched packages ✅ · check ✅ · license/cla ⏳ pending. Changelog ✅ — docs/changes/unreleased/1670-task-run-audit.md.

2. Review passes (three, separate comments) — Pass 1 — ARCHITECTURE ❌ absent · Pass 2 — SECURITY ❌ absent · Pass 3 — CODE EFFICIENCY & DESIGN ❌ absent — 0 of 3 posted.

3. Proof on screen — user-facing ✅ (the spend-limit ask card "Raise to $X" / "Stop for today" and the top-bar raise). Video/GIF embedded in the body ❌ missing — required before ready.

4. Tags — were absent → now bug · area:session · area:chat · sev:serious · milestone Reliable agent.

Actions: converted to draft until the bar holds; the three review passes are running and the capture is queued. Nothing merges before all four conditions hold.

_{drafted with CodeAF}