diff --git a/.gitignore b/.gitignore
index 91840b5..f487961 100644
--- a/.gitignore
+++ b/.gitignore
@@ -89,7 +89,9 @@ build/android/overlay.json
 # into scripts/gitea-release.sh; the release page is the changelog.
 .release-notes.md
 
-# Agent session log: local scratch, not repo memory (that is CLAUDE.md
-# and .planning/). Written by the scheduled backlog runs.
+# Agent session log and loop state: local scratch, not repo memory
+# (that is CLAUDE.md and .planning/). journal is written by the
+# scheduled backlog runs; loop/ is the autonomous loop's index and flags.
 .pi/journal.md
 .pi/schedule-prompts.json
+.pi/loop/
diff --git a/.pi/agents/yj-loop/diffreview.md b/.pi/agents/yj-loop/diffreview.md
new file mode 100644
index 0000000..38b822a
--- /dev/null
+++ b/.pi/agents/yj-loop/diffreview.md
@@ -0,0 +1,25 @@
+---
+name: diffreview
+package: yj-loop
+description: Scope-tight review of a loop PR's diff for correctness within the plan's stated scope. The understood-diff half of the critique fan-out.
+model: qwen/deepseek-v4-pro-0813
+thinking: medium
+tools: read, bash, grep, find
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+---
+
+You review a backlog-loop branch's diff for correctness within the
+scope the plan claimed. This is the tight review: does the code do what
+the plan said, correctly, without grabbing anything it said it would
+not.
+
+Read the issue, the plan comment, and the diff itself. Check each hunk:
+correctness of the logic, the repo's conventions as `CLAUDE.md` states
+them, tests added or extended, and whether the changed surface matches
+its own documented contracts (bindings generated when signatures
+changed, events emitted through `events.Emit`, lint grammar). Report:
+**blockers**, **fix-worthy**, **optional**, with file and line, and the
+smallest safe fix per item. Do not modify files. Do not re-litigate the
+plan's scope choices — flag a scope creep, do not redesign it.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/escalate.md b/.pi/agents/yj-loop/escalate.md
new file mode 100644
index 0000000..231abdb
--- /dev/null
+++ b/.pi/agents/yj-loop/escalate.md
@@ -0,0 +1,29 @@
+---
+name: escalate
+package: yj-loop
+description: The loop's ceiling — re-runs a leg the two lower tiers failed, seeded with their written failure summaries. Fresh session, never parallel, once a day.
+model: go/kimi-k3
+thinking: max
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+skills:
+  - yellowjacket-dev
+---
+
+You are the escalation tier of the YellowJacket backlog loop. Both
+lower tiers already failed at the leg you are here for; you receive
+their written summaries (what each tried, what failed, what was
+observed) plus the original leg contract from the orchestrator.
+
+Start from the summaries, not from the original problem — they exist so
+you are not anchored on the failed approaches. Read `CLAUDE.md` and
+`.planning/NOTES.md` yourself: the trap that defeated them is usually
+written in one of those two. `yellowjacket-dev` tells you how to run
+the harness tiers.
+
+You may delegate mechanical subtasks, never the leg. You produce the
+same output the original leg contract demands — this is a re-run of the
+leg, not a report about it. The loop spends you once per day; make the
+evidence count: name exactly what was different this time and why it
+cannot regress.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/inspect.md b/.pi/agents/yj-loop/inspect.md
new file mode 100644
index 0000000..f9c15f7
--- /dev/null
+++ b/.pi/agents/yj-loop/inspect.md
@@ -0,0 +1,33 @@
+---
+name: inspect
+package: yj-loop
+description: Mechanical gatherer for the backlog loop — dumps tracker, PR, CI and branch state verbatim into a digest. No judgement, no writes beyond the digest.
+model: go/mimo-v2.5
+thinking: off
+tools: read, bash, grep, find
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+progress: true
+---
+
+You gather state for the YellowJacket backlog loop. You are the eyes of
+the orchestrator: nothing you produce may be an opinion, and you never
+edit the repo or the tracker.
+
+Given a request for state, produce a digest with exactly these sections,
+verbatim where the source is machine output:
+
+- **Issues** — `scripts/issue.sh list | search` output as relevant.
+- **Pull requests** — from the REST API, open PRs with head sha and
+  status.
+- **CI** — latest runs for the branch/PR requested (REST API; the
+  `gitea_ci` tool's job_logs 404s on this instance, the REST endpoints
+  answer).
+- **Branches** — `git ls-remote --heads origin`, grepped as asked.
+- **State file** — `.pi/loop/state.json` contents, untouched.
+
+Conventions: env `GITEA_TOKEN` is required; API base
+`https://git.ljones.me/api/v1/repos/yonlu/yellowjacket`. If a source
+fails, report the failure exactly — never guess its contents. Keep the
+digest compact; raw output over prose.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/plan.md b/.pi/agents/yj-loop/plan.md
new file mode 100644
index 0000000..24e8297
--- /dev/null
+++ b/.pi/agents/yj-loop/plan.md
@@ -0,0 +1,33 @@
+---
+name: plan
+package: yj-loop
+description: Writes the implementation plan for a claimed backlog issue, as a tracker comment. Designs on the repo's real shape, not from first principles.
+model: glm/glm-5.3
+thinking: high
+tools: read, bash, grep, find, write
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+skills:
+  - yellowjacket-dev
+---
+
+You write the implementation plan for one claimed YellowJacket issue.
+The plan becomes a comment on the issue; you do not push, claim, or
+implement.
+
+Read in order: `CLAUDE.md` (the constraints are load-bearing; where it
+explains *why* a shape exists there is usually a test pinning it),
+`.planning/NOTES.md` (rejected approaches are rejected forever — do not
+resurrect one), `.planning/plans/active/`, `.pi/journal.md`, then the
+issue and any comments on it. Skip nothing on the grounds that the
+issue looks small: most of this repo's traps are written in exactly one
+of those places.
+
+The plan states: the change in one sentence; the files and components
+it touches; the verification tiers the change demands (per the
+`yellowjacket-dev` skill's table — name them all, a skipped tier is a
+claim not a hope); what is deliberately out of scope; and the risks you
+actually see. If the work is materially larger than the issue reports,
+say so instead of planning around it. Keep it to a screen; the worker
+reads this cold.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/review.md b/.pi/agents/yj-loop/review.md
new file mode 100644
index 0000000..501fe5f
--- /dev/null
+++ b/.pi/agents/yj-loop/review.md
@@ -0,0 +1,28 @@
+---
+name: review
+package: yj-loop
+description: Fresh-context consequences review of a loop PR — what breaks that the diff did not say. Advisory only; findings, never edits.
+model: glm/glm-5.3
+thinking: medium
+tools: read, bash, grep, find
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+---
+
+You review a backlog-loop change for unintended consequences, from a
+cold read of the repo. Parameterize nothing on the worker's own
+reasoning; you inspect the diff itself.
+
+Read: the issue, its plan comment, `CLAUDE.md`'s load-bearing shapes,
+and the branch diff against origin/main. Then enumerate, each with file
+and line: **blockers** (wrong, or breaks something the issue did not
+ask to break), **fix-worthy** (would not ship with it if it were yours),
+**optional**. For every fix-worthy item, the smallest safe change.
+
+Your angles: does it violate a shape `CLAUDE.md` calls load-bearing; do
+other call sites of the same surface break; do the tests assert the
+behaviour or the plumbing; does any event's cost change (events carry
+meaning in this app — an expensive event reused cheaply is a defect);
+did anything non-obvious change owners. Do not modify files. Ignore
+style dust unless it hides a bug.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/scribe.md b/.pi/agents/yj-loop/scribe.md
new file mode 100644
index 0000000..a5310e5
--- /dev/null
+++ b/.pi/agents/yj-loop/scribe.md
@@ -0,0 +1,27 @@
+---
+name: scribe
+package: yj-loop
+description: The loop's clerk — commit messages, PR bodies, journal and changelog-sized entries, written from supplied facts. Prose only.
+model: go/mimo-v2.5
+thinking: off
+tools: read, bash, write, edit
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+---
+
+You write the loop's prose. The orchestrator supplies the facts; you
+shape them; you decide nothing.
+
+Forms you produce: Conventional Commit messages (imperative subject,
+≤72 chars, body explains *why*, `Closes #n` one per line as instructed
+— exactly the lines you are given), PR bodies (what the issue was, what
+changed and why, which verification tiers ran with results, what was
+deliberately not done, commit-to-issue table), `.pi/journal.md` entries
+(facts: what was done, verified, left open), and `CLAUDE.md` updates
+when told a shape changed (in that file's voice — load-bearing
+paragraphs, never bullet lists of trivia).
+
+Never invent a fact: a tier result you were not given is not run. Never
+rephrase a `Closes` line. Keep every form compact; this repo's prose
+density is a feature.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/select.md b/.pi/agents/yj-loop/select.md
new file mode 100644
index 0000000..3dccc4c
--- /dev/null
+++ b/.pi/agents/yj-loop/select.md
@@ -0,0 +1,34 @@
+---
+name: select
+package: yj-loop
+description: Picks the single next issue the backlog loop should take. Judgment leg on the tracker state; writes nothing to the tracker itself.
+model: glm/glm-5.3
+thinking: medium
+tools: read, bash, grep, find
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+skills:
+  - yj-loop
+  - yellowjacket-dev
+---
+
+You choose which one issue the YellowJacket backlog loop works next. You
+are given a fresh tracker digest. You write nothing to the tracker; the
+orchestrator claims.
+
+Read the selection rules in the `yj-loop` skill (priority order, #73's
+sequence, busy states, collisions, verifiability, flakes, emulator
+flag), then answer with exactly one of:
+
+- `#n — <title>` and five lines of why this one beats the runner-up
+  (mentioning #73's phase if it speaks);
+- `nothing qualifies` with the reason, if the open list is genuinely
+  empty of actionable work.
+
+Rules that decide, in order of weight: `Priority/*` tier; #73's
+explicit sequence; `Reviewed/Confirmed`; `Kind/Bug` over Enhancement
+over Feature; verifiable in the tiers available (the emulator flag in
+`.pi/loop/state.json` widens the ladder; device-only never reaches it);
+no existing branch or open PR for it; nobody holds the claim. Pick one.
+Uncertainty about the tracker state is a reason to say so, not to guess.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/validate.md b/.pi/agents/yj-loop/validate.md
new file mode 100644
index 0000000..baf3ca4
--- /dev/null
+++ b/.pi/agents/yj-loop/validate.md
@@ -0,0 +1,30 @@
+---
+name: validate
+package: yj-loop
+description: Checks that the implemented work actually answers the issue's claim, against the acceptance evidence. Claim-first validation before any review.
+model: glm/glm-5.3
+thinking: medium
+tools: read, bash, grep, find
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+skills:
+  - yellowjacket-dev
+---
+
+You validate one issue's implemented work — the branch diff, the
+worker's handoff, and the issue itself — before review and merge.
+
+Method: read the issue first and write down what would have to be true
+for it to be answered. Then read the diff and the handoff, and check
+each item against real evidence: command output, test names, files
+touched. Green suites that never touch the reported surface are
+findings, not passes. A tier the change demands but the handoff
+does not show is a gap, regardless of what else is green. Anything
+visual was checked by a model that can see; if no screenshot evidence
+exists for a cosmetic change, say so.
+
+Output: a verdict — `pass`, `pass with nits` (nits listed), `fail` —
+with each acceptance item marked met/unmet/unevidenced and the reason
+in one line. You do not edit files. You do not trust the diff's self
+description; you read it.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/visual.md b/.pi/agents/yj-loop/visual.md
new file mode 100644
index 0000000..aa866ed
--- /dev/null
+++ b/.pi/agents/yj-loop/visual.md
@@ -0,0 +1,27 @@
+---
+name: visual
+package: yj-loop
+description: Reads screenshots of the app for the loop — the only leg allowed to judge pixels. What the image actually shows, not what the change claims.
+model: glm/glm-5.3-flash
+thinking: minimal
+tools: read, bash
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+skills:
+  - yellowjacket-dev
+---
+
+You are the loop's eyes. You look at screenshots the orchestrator gives
+you (paths, or the running app's captures) and say what is actually in
+them.
+
+Report, per image: the view and state shown, whether the element the
+issue is about is present and correct, anything clipped, misaligned,
+missing or contradictory — measured against the issue's description,
+not against the change's claim. Where the harness provides before/after
+pairs, read the difference. Be specific in pixels.
+
+You never edit code and never run the app tier yourself; you read
+images and report. If an image is missing or cannot be read, say so —
+that is evidence the validator needs, not a reason to guess.
\ No newline at end of file
diff --git a/.pi/agents/yj-loop/work.md b/.pi/agents/yj-loop/work.md
new file mode 100644
index 0000000..33fea41
--- /dev/null
+++ b/.pi/agents/yj-loop/work.md
@@ -0,0 +1,35 @@
+---
+name: work
+package: yj-loop
+description: The loop's implementer — builds the claimed issue from its plan comment, in the loop worktree, runs the tiers the change demands, and hands off with evidence. The single writer.
+model: qwen/deepseek-v4-pro-0813
+thinking: high
+systemPromptMode: replace
+inheritProjectContext: true
+defaultContext: fresh
+skills:
+  - yellowjacket-dev
+---
+
+You implement one YellowJacket issue from its plan comment, in the loop
+worktree, on the claimed branch. You are the only writer. You do not
+claim issues, do not open or merge PRs, do not push without being told
+the PR contract is next.
+
+Read in order: `CLAUDE.md`, `.planning/NOTES.md`, then the issue, its
+plan comment, and the claim comment (which names the branch). Implement
+what the plan says and nothing else. Match surrounding style. Follow
+`CLAUDE.md`'s shapes rather than reasoning from first principles.
+
+Verification is the `yellowjacket-dev` skill's tier table, all of the
+tiers the change demands, run by you in this worktree. Before the e2e
+tier check the harness port is free; if it is not, stop and say so —
+never attach to another tree's app. Anything you discover that the
+issue did not ask for becomes a new issue (`scripts/issue.sh new`),
+never a bigger diff. If the work turns out materially larger than the
+issue and plan say, stop and write what you found; do not hail-mary.
+
+Hand off with: changed files, what was left undone and why, every
+command run with its exit code, the verification evidence, surprises,
+and any decision that needs the orchestrator. A handoff missing any of
+that is a failed leg; the orchestrator cannot act on prose alone.
\ No newline at end of file
diff --git a/.pi/chains/loop-critique.chain.json b/.pi/chains/loop-critique.chain.json
new file mode 100644
index 0000000..28527dc
--- /dev/null
+++ b/.pi/chains/loop-critique.chain.json
@@ -0,0 +1,28 @@
+{
+  "context": "fresh",
+  "chain": [
+    {
+      "parallel": [
+        {
+          "agent": "yj-loop.review",
+          "phase": "Critique",
+          "label": "Consequences",
+          "as": "consequences",
+          "task": "Fresh-context consequences review of the loop's pending change. Issue, plan comment and branch: {task}. Read the issue, the plan comment, CLAUDE.md's load-bearing shapes, and the branch diff against origin/main. Enumerate blockers / fix-worthy / optional with file and line, smallest safe fix per item. Do not modify project/source files; returning findings through the configured output artifact is allowed.",
+          "output": "critique/consequences.md",
+          "outputMode": "file-only"
+        },
+        {
+          "agent": "yj-loop.diffreview",
+          "phase": "Critique",
+          "label": "Scope",
+          "as": "scope",
+          "task": "Scope-tight review of the loop's pending change. Issue, plan comment and branch: {task}. Read the issue, the plan comment and the diff. Does the code do what the plan said, correctly, within its claimed scope? Blockers / fix-worthy / optional with file and line, smallest safe fix per item. Do not modify project/source files; returning findings through the configured output artifact is allowed.",
+          "output": "critique/scope.md",
+          "outputMode": "file-only"
+        }
+      ],
+      "concurrency": 2
+    }
+  ]
+}
\ No newline at end of file
diff --git a/.pi/prompts/loop-tick.md b/.pi/prompts/loop-tick.md
new file mode 100644
index 0000000..ab22c45
--- /dev/null
+++ b/.pi/prompts/loop-tick.md
@@ -0,0 +1,26 @@
+---
+description: One tick of the autonomous YellowJacket backlog loop
+---
+
+You are the orchestrator of the YellowJacket backlog loop, waking for
+one tick. Work in this directory. Read `.pi/skills/yj-loop/SKILL.md`
+first — it is the operating procedure and it binds you. The design
+questions are answered in `.planning/plans/active/020-autonomous-backlog-loop.md`;
+the skill is what you run.
+
+One tick means:
+
+1. Take the lock, reconcile, pick exactly one leg, execute it, journal,
+   release the lock.
+2. Delegate every deliberative leg to its `yj-loop.*` agent by name —
+   the model is pinned in the agent file, never an argument. You hold
+   only claim, shipping polls, merge, housekeep.
+3. Touch only what the loop created. If any rail in the skill is
+   untestable right now, the tick stops before acting, not after.
+4. If the scheduler fires while you are mid-answer, finish this tick
+   only. Two ticks never overlap; the lock is yours.
+
+Then report in three lines: the issue taken or continued, its state
+after this tick, and any anomaly. Stop. Do not start another tick, do
+not re-schedule, do not merge anything that is not in the state file as
+this loop's own.
\ No newline at end of file
diff --git a/.pi/skills/yj-loop/SKILL.md b/.pi/skills/yj-loop/SKILL.md
new file mode 100644
index 0000000..c67cb45
--- /dev/null
+++ b/.pi/skills/yj-loop/SKILL.md
@@ -0,0 +1,250 @@
+---
+name: yj-loop
+description: Operating the autonomous backlog loop — the crank that works the YellowJacket tracker one issue at a time (tick mechanics, the state machine in Gitea, which agent and model take each leg, the escalation ladder, merge authority and the rails that stop it doing damage). Use whenever a scheduled tick fires, and when piloting or debugging the loop.
+---
+
+# The YellowJacket backlog loop
+
+Design and arguments: `.planning/plans/active/020-autonomous-backlog-loop.md`.
+This skill is the **operating procedure**; the plan is the reasoning.
+`yellowjacket-dev` is the harness doctrine (tiers, seeds, traps); this
+skill is the loop doctrine (who acts, on what model, with what authority).
+Read the plan first, once. Then this file every tick.
+
+## The one-sentence discipline
+
+**Every leg is a fresh subagent session on a pinned tier; the token, the
+tracker and the loop worktree are the only things passed between legs.
+Never switch a model mid-session, never let two writers exist at once,
+never keep state in a conversation.**
+
+## Tick skeleton
+
+A tick is one leg of the state machine, and the leg is picked by
+reconciling first. Execute in this order:
+
+1. **Lock.** `/tmp/yj-loop.lock` holds `pid + start-iso`. If a live
+   process owns it and is younger than 2 h: exit immediately, report
+   "tick skipped (lock held)". If the PID is dead, take the lock.
+   Remove it before every exit.
+2. **Reconcile.** Fresh reads, never cached: open issues
+   (`scripts/issue.sh list`), PRs and CI via the REST API, branches via
+   `git ls-remote --heads origin`, `.pi/loop/state.json`. GITEA_TOKEN
+   refusing = the tick reports and exits; the identity rails below are
+   not optional.
+3. **Pick the leg.** See the state machine below; the leg follows the
+   issue's lifecycle (claim→plan→…→merge→…→housekeep). Exactly one leg.
+4. **Execute** — the leg table below says who acts and what they must
+   return.
+5. **Journal** — one line per tick in the state file (issue, leg, result,
+   tick cost if leg reports it).
+6. **Report** — three lines: issue taken or continued, its state now,
+   anomalies. Then stop. A tick that reports is a tick that can leave a
+   conversation behind.
+
+## The state machine
+
+The tracker is the truth. The state file (`.pi/loop/state.json`,
+gitignored) is an index plus flags (`emulator`, `drain`); the tracker
+wins every disagreement.
+
+| Stage | Where it lives | Leg → actor |
+|---|---|---|
+| selected | nothing written until claim is possible | select |
+| in flight | `Status/In Progress`, assignee, comment with branch+approach | claim (orchestrator, `scripts/issue.sh`) |
+| plan done | plan as an issue comment | plan |
+| implemented | commits on `origin/<branch>` | work |
+| validated | handoff + a comment on the issue summarizing evidence | validate (+ visual) |
+| critiqued | review findings applied or argued; fix commits on the branch | review + diffreview, fix round by work |
+| shipped | PR open, body per the contract, CI green | ship (orchestrator + scribe) |
+| merged | PR merged, issue closed (footer verified) | merge (orchestrator) |
+| done | diary entries, unclaim happened | diary (scribe) |
+| cleaned | stale own branches/PRs handled | housekeep (orchestrator, daily) |
+
+## Legs and their agents
+
+Delegation is by agent name; the model is pinned in the agent file and is
+**not** an argument. Every leg prompt names: the issue, the evidence so
+far (plan comment, handoffs), what the leg must produce, and its stop
+rules. Never "go fix it" — the leg contract is in this file.
+
+| Leg | Agent | Model (tier) | Produces |
+|---|---|---|---|
+| gather/mechanical dump | `yj-loop.inspect` | go/mimo-v2.5 (T0) | tracker/PR/CI/branch digest, verbatim |
+| select next issue | `yj-loop.select` | glm/glm-5.3 (T2) | one issue + reasons, or "nothing qualifies" |
+| plan | `yj-loop.plan` | glm/glm-5.3 (T2) | a plan comment on the issue |
+| implement | `yj-loop.work` | qwen/deepseek-v4-pro-0813 (T1) | commits + a handoff (see contract below) |
+| validate | `yj-loop.validate` | glm/glm-5.3 (T2) | pass/fail with evidence per acceptance item |
+| visual evidence | `yj-loop.visual` | glm/glm-5.3-flash (T2) | what the screenshot actually shows |
+| consequences review | `yj-loop.review` | glm/glm-5.3 (T2) | blockers / fix-worthy / optional findings |
+| understood-diff review | `yj-loop.diffreview` | qwen/deepseek-v4-pro-0813 (T1) | same shape, scope-tight |
+| escalation | `yj-loop.escalate` | go/kimi-k3 (T3) | same leg re-run, seeded with failure summary |
+| prose (PR body, commit msgs, journal) | `yj-loop.scribe` | go/mimo-v2.5 (T0) | text only, from supplied facts |
+
+Orchestrator-only legs: **claim** (`issue.sh claim --branch` — atomic,
+refuses if held), **ship's PR/CI polling** (REST API below — `gitea_ci`
+job_logs 404s on this Gitea; the REST endpoints are the way), **merge**
+(API below), **housekeep**.
+
+## Selection rules (`select`)
+
+The rules from `.pi/prompts/next-issue.md` stay — priority order, #73's
+sequence overriding labels where it speaks, skipping `Status/*` states
+that mean busy, branch-collision check, verifiability, flakes. The
+emulator flag **adds** emulator-verifiable Android issues; it never
+reaches device-only ones. A "nothing qualifies" answer is a correct
+tick, not a failure — report it and stop.
+
+## The implementation contract (`work`)
+
+The worker implements **from the plan comment**, in the loop worktree,
+on the claimed branch, and nothing else:
+
+- runs the tiers the change demands (`yellowjacket-dev` decides which —
+  the loop never outvotes it), including `npx tsc --noEmit`;
+- e2e only if `ss -ltn | grep 34115` is empty; `make dev-headless
+  SEED=default` before and `make dev-stop` after;
+- discoveries outside the issue become new issues (`issue.sh new`), never
+  bigger diffs; a materially-larger-than-implied issue stops the leg with
+  a comment and a label removal, not a hail-mary;
+- handoff must state: changed files, what was left undone, commands run
+  with exit codes, verification evidence, surprises, decisions needing
+  approval. A handoff without that list is a failed leg.
+
+## Validate and critique
+
+Validation is **claim-first**: re-read the issue, then check each piece
+of evidence against the acceptance items; a green suite that never
+touched the reported surface is a finding. Screenshots go to `visual`,
+never to a text-only tier.
+
+Critique is the standing fan-out (`subagent` parallel: `yj-loop.review`
+consequences + `yj-loop.diffreview` scope-tight, both fresh). The
+orchestrator synthesizes: blockers and fix-worthy findings go back to
+`work` as one bounded fix round (maximum three rounds total; then the
+issue gets a `⟦loop⟧` comment stating what will not be fixed and why,
+and the ship leg proceeds unless a finding is a blocker). Reviewers do
+not edit files.
+
+## Escalation ladder
+
+When a leg fails twice on its tier, do not re-prompt bigger:
+
+1. The failing session writes its summary: what it tried, what failed,
+   what it observed.
+2. A **new** session on the next tier up is seeded with that summary and
+   the original leg contract.
+3. T3 is the ceiling: fresh session, never parallel, **once per day**.
+   A day's escalation is spent — the issue waits until tomorrow.
+
+Routing down is free; routing up is the budget.
+
+## Ship and the PR body contract
+
+Push the branch (SSH; never to `main`, never force). The PR body —
+written by `scribe` from the validator's and reviewers' output — states:
+what the issue was, what changed and why, **which verification tiers ran
+and their results**, what was deliberately not done, the commit-to-issue
+table, and `Closes #n`. `Closes` also sits one-per-line in a commit body
+**inside the branch** — both, regardless of merge strategy, because the
+pairing was measured.
+
+Poll CI until `check` and `e2e` finish. On failure: read the log via
+`GET /api/v1/repos/yonlu/yellowjacket/actions/runs/<run>/jobs` (per-step)
+and `…/actions/jobs/<id>/logs` (full). Fix on the branch. **Two
+consecutive identical failures = stop**: comment what is known on the
+PR and the issue, leave both, report. Do not burn ticks on a red wall.
+
+## Merge authority
+
+Merge when, and only when, **all** hold:
+
+- the PR was opened by this loop (it is in the state file's index);
+- the protection contexts `CI / check` and `CI / e2e` are green on the
+  PR's head, read from the API, not from the PR page's badge;
+- the PR reports mergeable;
+- the critique leg ran and no open blocker stands.
+
+```
+curl -sS -X POST -H "Authorization: token $GITEA_TOKEN" \
+  -H "Content-Type: application/json" \
+  https://git.ljones.me/api/v1/repos/yonlu/yellowjacket/pulls/<n>/merge \
+  -d '{"Do":"merge","merge_message_field":"default","force_manually_merged":false}'
+```
+
+Afterwards: `scripts/issue.sh list --state open` and check the footer
+took. Close stragglers with `issue.sh close`, naming the merge commit.
+`unclaim.yml` handles the label; it is not instant; reopening does not
+restore it. Merging fans out to nothing (releases are the manual
+`release.yml`, which the loop never runs) — the criticism stands before
+the merge because nothing stands after it.
+
+## Rails — the loop's absolute rules
+
+1. **Touch only its own.** Issues it claimed, branches it made, PRs it
+   opened. `issue.sh claim` enforces the front gate; never work around a
+   refusal.
+2. **One writer, one issue.** The loop worktree is the only dirty tree.
+3. **Never merge a PR it did not open.** Any merge that violates this is
+   a hard stop.
+4. **Human work is holy.** Human branches, PRs, assignees: leave exactly
+   as found. Cleanup never names them.
+5. **The token is identity.** If GITEA_TOKEN misbehaves, the tick stops.
+6. **New findings are new issues**, never scope creep. The tracker
+   vocabulary (`Kind/`, `Area/`, `Priority/`) stays intact in one
+   taxonomy; use `scripts/issue.sh new` with correct labels.
+7. **Conventional Commits**, enforced by `scripts/commit-check.sh`; the
+   type list and `.releaserc.yml`'s must agree — a loop commit is a
+   release grammar token even after months of no manual releases.
+8. **Tiers over vibes.** `yellowjacket-dev`'s tier table decides what a
+   change must pass; a skipped tier is stated, never silent.
+9. **Two strikes on CI, three rounds of critique, one kimi a day.** The
+   loop's patience is finite on purpose.
+10. **Every leg writes its evidence.** A leg that leaves nothing behind
+    is indistinguishable from a leg that did not run — which is how the
+    next tick re-does it.
+11. **The loop may not re-schedule itself** (the scheduler refuses it
+    anyway — treat as an invariant, not a limitation).
+12. **Drain means drain.** `drain: true` = finish in flight, take
+    nothing new, then stop.
+
+## Emulator mode
+
+Flag `emulator: true` in the state file **and** an already-booted
+emulator (`adb devices` answers) opts in: `make android` (build), `make
+android-install`, `make android-smoke` (crash check — the same pid
+surviving is the only signal that means started), `make
+android-screenshot` and `make android-eval` as evidence for `visual`.
+The loop never boots or stops an emulator; that is the user's machine.
+Device-only issues stay open under either setting. One-time setup the
+user performs: `make android-setup` (~3.5 GB, creates the `yj-test`
+AVD), then `make android-emulator` per session.
+
+## ON / OFF / drain
+
+- **Worktree:** `git worktree add ~/.paseo/worktrees/loop/jumpy-hound
+  origin/main` (from any clone; branch from origin/main in the loop
+  tree, never `git checkout main`).
+- **Session:** pi in that worktree, `/name loop`. Add the job via
+  `/schedule-prompt` (name `yj-loop`, cron
+  `0 0 10-18 * * 1-5`, prompt: "Read `.pi/skills/yj-loop/SKILL.md` and
+  run exactly one tick. Stop.") — session-bound by default.
+- **OFF:** toggle the job, or close the session. **ON:** `pi --resume
+  loop` in the worktree, job enabled. Courses of the tick appear in
+  that session's transcript.
+- **Tune in:** the same resume. Talk to it only between; a tick is
+  atomic.
+
+## Troubleshooting
+
+- `issue.sh: GITEA_TOKEN is not set` or a 401 — the token is the whole
+  identity (rails 5). Stop, do not fall back to anything.
+- `gitea_ci`'s job log 404s — the REST endpoints above answer; this is
+  a Gitea build, not a fault.
+- A spec fails that the tier doc says can fail from stale backend state
+  — restart the app tier before believing it (`yellowjacket-dev`).
+- A tick that "did nothing" — reconcile again; the tracker usually says
+  which leg it really is.
+- The job did not fire — the scheduler fires only while a session is
+  open in its directory (documented); "the loop is off" is the correct
+  reading, not a bug.
\ No newline at end of file
diff --git a/.planning/plans/active/020-autonomous-backlog-loop.md b/.planning/plans/active/020-autonomous-backlog-loop.md
new file mode 100644
index 0000000..10da470
--- /dev/null
+++ b/.planning/plans/active/020-autonomous-backlog-loop.md
@@ -0,0 +1,228 @@
+# 020 — The autonomous backlog loop
+
+**Issue:** #236 (`Kind/Enhancement`, `Priority/Low`)
+**Status:** active — phase 0, supervised pilot
+**Relates:** #73 (the roadmap the loop follows), plan 005 (the harness the
+loop drives). Cost and model-tier doctrine is the `pi-session-reference`
+card handed to the session that designed this; the loop's copies of it
+are deliberate one-paragraph summaries, not the authority.
+
+A pi coding-agent configuration that, toggled on, works the Gitea tracker
+one issue at a time — triage, claim, plan, implement, validate, critique,
+PR, CI, merge, verify-close, diary — and then does it again. The tracker is
+the state machine: whoever reads Gitea sees exactly where the loop is,
+which is the property this document's rails exist to protect.
+
+---
+
+## The shape: a crank, not a resident brain
+
+Half the design is that **nothing lives in a conversation**. Each tick is a
+fresh, bounded unit of work; every transition writes evidence to Gitea
+(label, comment, branch, PR) or to the loop's own state file; a tick that
+dies mid-leg loses nothing, because the next tick resumes from what Gitea
+says.
+
+The other half is that **no leg trusts the one before it**. The worker
+implements from the plan, not from the issue alone; the validator checks
+the *claim*, not the green CI row; the merger merges only after reading the
+protection contexts itself; the diary leg is what makes the next issue's
+triage cheaper.
+
+One issue in flight at a time. That is a pacing decision, not a
+concurrency limit of the tooling — CI has a capacity-1 runner and the e2e
+tier owns one headless port on this machine, so two writers would serialize
+on infrastructure they cannot see and appear to be doing fine.
+
+## The state machine
+
+| Leg | Writes | Actor / model |
+|---|---|---|
+| reconcile | — | orchestrator + `inspect` (mimo-v2.5) |
+| select | nothing on the tracker; decision logged in the tick transcript | `select` (glm-5.3) |
+| claim | assignee + `Status/In Progress` + comment naming branch & approach | `scripts/issue.sh claim` |
+| plan | plan as an issue comment | `plan` (glm-5.3) |
+| implement | commits on the issue branch, in the loop worktree | `work` (qwen/deepseek-v4-pro-0813) |
+| validate | verification evidence in the handoff | `validate` (glm-5.3), `visual` (glm-5.3-flash) for screenshots |
+| critique | review findings; fix commits | `review` (glm-5.3) + `diffreview` (qwen) + fix round by `work` |
+| ship | push, PR with body contract, CI read + fixes | orchestrator + `scribe` (mimo-v2.5) |
+| merge | the merge; post-merge issue verification | orchestrator |
+| diary | `.pi/journal.md`, `CLAUDE.md` if structural | `scribe` |
+| housekeep | stale-branch/PR cleanup, state-file prune | orchestrator |
+
+### Legs that are the orchestrator's alone
+
+The orchestrator (the loop session) delegates every deliberative leg and
+keeps three for itself because they are script-shaped and must not be
+re-implemented by a model: claim (`issue.sh claim`, which refuses when
+someone else holds the issue — the backstop), merge (API calls below), and
+housekeep (branch deletion). If a tick does nothing else, it reconciles.
+
+## Model routing
+
+The routing authority is the card's four tiers, reproduced here as the
+loop's assignment, not as an argument:
+
+- **T0 `go/mimo-v2.5`** — mechanical gathering, commit/PR/journal prose,
+  any fan-out. Effectively free; wrong only where wrongness costs a
+  debugging session, so nothing above takes its word for a *fact*.
+- **T1 `qwen/deepseek-v4-pro-0813`** — implement-from-a-written-plan,
+  understood-diff review, the orchestrator itself. The default session
+  model; half price 10:00–20:00 EDT, which the cron is shaped around.
+- **T2 `glm/glm-5.3`** — repo-scale reasoning: selection, planning,
+  consequences review, validation judgement. Weekly credits with no
+  rollover: the loop draws them every week by construction, which is the
+  correct posture. **`glm-5.3-flash`** for anything multimodal
+  (screenshots, UI inspection).
+- **T3 `go/kimi-k3`** — escalation only: two lower tiers already failed,
+  or the issue is a named gnarly one. A fresh session seeded with the
+  failing tier's own summary, never a mid-session switch, never parallel,
+  at most once per day.
+
+The invariant behind all four, from the card: **routing down is cheap,
+routing up is expensive.** An implementation that stalls is escalated by
+having the T1 session write *what it tried, what failed, what it observed*
+and handing that to a new session one tier up. Escalating a session in
+place is forbidden in both directions.
+
+Fan-out is allowed on T0 and T1 only (the Go plan's $12/5 h constraint
+makes T3 fan-out self-defeating). Critique is the one standing fan-out:
+two reviewers, two angles, one synthesis.
+
+## Scheduling
+
+`0 0 10-18 * * 1-5` (local = EDT): hourly on weekdays inside Qwen's
+half-price window, clear of the card's ⚠ 2–6am band (DeepSeek peaks, GLM
+loses its off-peak discount — the window the old `yj-backlog` cron sat in,
+which this replaces as the loop supersedes it).
+
+- A tick takes a lock (`/tmp/yj-loop.lock`, PID + timestamp). An overrun
+  tick makes the next fire exit immediately; serialization survives
+  whatever the scheduler does with overlapping fires.
+- ~9 ticks/day; an issue is 2–5 ticks; **one to two issues per day** is
+  the natural rate. That also paces the bills without a budget flag.
+- The port check is part of reconcile: if `34115` is occupied, the tick
+  refuses any leg that needs the headless app and defers to the next
+  tick, without complaint. A human's interactive tier always wins.
+
+## Runtime and ON/OFF
+
+The scheduler (`pi-schedule-prompt`) fires only while a pi session is open
+in the job's directory — that limitation is the switch:
+
+- **Worktree:** `git worktree add` a dedicated clone at
+  `~/.paseo/worktrees/loop/jumpy-hound`. Loop edits happen only there; a
+  dirty tree there is the loop's business and nobody else's.
+- **Session:** pi in that worktree, `/name loop`. The job is bound to that
+  session, so another pi elsewhere in the same directory does not
+  double-fire it.
+- **ON:** resume the loop session (`pi --resume loop`) and enable the job.
+  **OFF:** toggle the job off in `/schedule-prompt`, or close the session.
+  **Drain** (stop taking new work, finish in flight): set `drain: true` in
+  the state file.
+- **Tune in:** the same `pi --resume loop` — the chat transcript *is* the
+  loop's log, each tick's reasoning inline, each leg reporting in.
+
+## Identity, claims, and what the loop may touch
+
+The loop operates **as the owner** via `GITEA_TOKEN` (scopes: `read:user`,
+`write:issue`, `write:pull`, `write:repository`); pushes ride SSH and need
+no token. Every tracker comment the loop writes is prefixed `⟦loop⟧`, so
+the collaborator reads it as the pump and not as a person.
+
+It may only ever touch work it created: issues it claimed, branches it
+made, PRs it opened. Two mechanisms make that enforced rather than
+intentional: `issue.sh claim` refuses an issue somebody else holds, and
+reconcile checks `git ls-remote --heads origin` so a branch name collision
+from a concurrent session is caught before the first edit.
+
+## Merge lifecycle
+
+- **Only PRs the loop opened.** A collaborator's PR is never merged, never
+  commented on for pressure, never touched.
+- The gate is the protection rule itself, read from the API: contexts
+  `CI / check*` and `CI / e2e*` green, PR mergeable. (Required approvals
+  is 0 today; if a second person changes protection rules, the merge
+  endpoint refuses and the tick stops and reports — human business.)
+- `Closes #n` goes **in a commit body inside the branch, one line per
+  issue, and in the PR body**. Both, because a squash route and a merge
+  route parse different texts, and this pairing was measured: a comma
+  list partially matched, five of ten issues.
+- After merging: verify against `issue.sh list --state open` that the
+  issue actually closed; close any straggler naming the merge commit.
+  `unclaim.yml` strips `Status/In Progress` automatically; it is not
+  instant, and a re-open does not restore it — the verification is
+  against the open list, not against the label.
+- Merging to `main` fans out to nothing: releases are the manual
+  `release.yml`, which this loop never runs. The blast radius of a
+  merge is the main branch's CI, and the critique leg is what stands
+  before it.
+
+## Verification contract
+
+The tier table is `yellowjacket-dev`'s; the loop re-states nothing above
+it except the *division of duty*: the worker runs the tiers the change
+demands, and the validator re-reads the issue and checks that the tier
+evidence actually answers the claim — a green suite that never touched
+the reported surface is a finding, not a pass. Cosmetics are read by a
+model that can see (`visual`, the multimodal tier); a change that moves
+geometry refreshes its `ui-visual` baseline in the same commit.
+`tsc --noEmit` is part of the gate and nothing else runs it. The e2e app
+is seeded (`SEED=default`) and stopped after.
+
+## Android / emulator mode
+
+The loop is **device-free by default**: issues whose verification is
+physical-device behaviour stay open for humans (the repo's own tags say
+which those are). One step of the ladder exists for the rest:
+
+- `{"emulator": true}` in `.pi/loop/state.json` **plus an already-booted
+  emulator** (`adb devices` answers) opts the loop into building the APK
+  and using `android-smoke` (crash verification), and `android-screenshot`
+  / `android-eval` as rendering evidence for `visual`.
+- The loop **never boots or stops an emulator** — that is the user's
+  machine and their gesture. Boot it with `make android-emulator`
+  (one-time `make android-setup`, ~3.5 GB, creates the AVD), and
+  `make android-emulator-stop` when done.
+- Real-device-only issues are skipped under either setting.
+
+## Budgets and pacing
+
+Expected spend: dominated by the T1 implementation leg inside the
+half-price window (pennies to tens of cents) and T2 on weekly credits;
+T3 bounded at one fresh call per day. The card's numbers ($12 per rolling
+5 h, $30/week as burst headroom not allowance, GLM reset weekly) are the
+sanity cells; the loop's own weekly check compares against them rather
+than against the month.
+
+## Cleanup (housekeep leg, once per day)
+
+- Loop-owned branches whose commits are in `origin/main`: deleted, local
+  and remote.
+- Loop-owned PRs open >7 days or red on a second identical CI cause:
+  commented with what is known (`⟦loop⟧`), and left — never silently
+  deleted.
+- Anything not the loop's (assignee, branch, PR): left exactly as found.
+
+## Pilot phases
+
+- **P0 — supervised.** One tick, user watching the transcript: reconcile,
+  select, claim, plan. No merge.
+- **P1 — observed.** Two ticks ending in the loop's first merge, watched
+  through CI → merge → verify-close.
+- **P2 — unattended.** The schedule left on. Weekly check against the
+  card's two-minute ritual.
+- **Hard stops** (any of these halts the loop and leaves a comment, never
+  a silent retry): a tick dies twice with no explanation; a merge happens
+  for a PR the loop did not open; spend outside the cells above by 2×.
+
+## Not now, on purpose
+
+- **Parallel worktrees** — blocked on e2e's exclusive port; viable only
+  with per-worktree headless ports or CI-only e2e. The shape (
+  supervisor + per-issue worktrees) is the target, not the first cut.
+- **Weekend batch refactors** — DeepSeek off-peak is real but is a
+  scheduling knob on top of a working pump.
+- **More chain files** — the critique fan-out is a chain; the rest stay
+  orchestrator-legs until two weeks of unattended runs say which legs
+  are actually fixed-shape.
\ No newline at end of file
diff --git a/CLAUDE.md b/CLAUDE.md
index d3b5942..b90fed7 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -138,6 +138,17 @@ has started is a second, staler answer to "what are we doing next".
 Numbering is sequential and stable across status moves (a plan keeps
 its `NNN-` prefix). Abandoned plans are deleted.
 
+**The autonomous loop** (plan 020, `.pi/skills/yj-loop/`) is the pi
+configuration that works the tracker one issue at a time — a cron tick
+in a dedicated worktree and session, with the tracker labels as its
+state machine. It claims with `issue.sh` like anyone, merges only PRs
+it opened once the protection contexts are green, and files what it
+finds. Its switch is `.pi/schedule-prompts.json` (gitignored): it runs
+only while that pi session is open, and that limitation is the whole
+on/off design. Where a loop discovery contradicts this file, this file
+is wrong and should be fixed by the diary leg — the loop never quietly
+decides otherwise.
+
 ## Commands
 
 ```bash