feat(workflows): audit-warranted pre-step + priority and tag enforcement

Two changes land together because each is broken without the other.

open-tasks.org gains a new Phase A.1, evaluated only in Next Mode. The
phase reads :LAST_AUDIT: from notes.org and walks five state signals
(reminder/task mismatch, passed scheduled date, "waiting on X" matches a
shipped X, dead file: link, sub-task >75% DONE coverage). If the temporal
threshold of 14 days trips, or any signal fires, Next Mode offers a
task-audit run before producing the recommendation. Item 1 in the offer
is "run task-audit first" per the recommendation-at-item-1 convention.

task-audit.org gains two pieces. Phase C now enforces priority and
type-tag presence per the project's legend, applies the [#A] dating rule
from that legend, and re-assesses :quick: and :solo: from reconciled
facts. Unambiguous calls land autonomously. Ambiguous ones flag
NEEDS-USER instead of being guessed. A new Phase E stamps :LAST_AUDIT:
on completion.

4 files changed, 158 insertions, 20 deletions

diff --git a/.ai/workflows/open-tasks.org b/.ai/workflows/open-tasks.org
index b07f352..e90bc12 100644
--- a/.ai/workflows/open-tasks.org
+++ b/.ai/workflows/open-tasks.org
@@ -25,13 +25,51 @@ Don't route "task review" / "review tasks" here — those trigger the hygiene ha
 
 Issue all source reads as a single batch of parallel tool calls. They're independent:
 
-1. Read =notes.org= — sections *Active Reminders* and *Pending Decisions* only.
+1. Read =notes.org= — sections *Active Reminders*, *Pending Decisions*, and *Workflow State* (note the =:LAST_AUDIT:= date if present; "unset" or absent is treated as "never run").
 2. Read each of the 2-3 most recent files in =.ai/sessions/= — only the *Next Steps* subsection under =* Summary= (those capture recent "next time, do X" items that may still be open).
 3. Read =todo.org= — entries under the open-work header (=* $Project Open Work=, e.g. =* Homelab Open Work=). Skip the resolved header (=* $Project Resolved=).
 4. Run =date= for an accurate timestamp (used to evaluate deadlines + flag suspected completions).
 
 The phases below work entirely from this in-memory snapshot. The original split-workflow versions issued these reads sequentially in different files; consolidating saves round-trips and prevents the two implementations drifting apart.
 
+* Phase A.1: Evaluate audit-warranted (Next Mode only)
+
+Before Next Mode produces a recommendation, decide whether a =task-audit= run looks warranted. List Mode skips this phase — it displays the plate as-is and should not trigger a write-side-effect workflow first.
+
+Two triggers; either one trips the warranted flag.
+
+** Temporal threshold
+
+If =:LAST_AUDIT:= is absent / unset, or its date is older than 14 days from today, the temporal trigger trips. (The 14-day default is per-project; tune by editing this workflow if a project churns faster or slower.)
+
+** State-signal scan
+
+Walk the Phase A snapshot for any of these cheap signals — one is enough:
+
+1. *Reminder/task mismatch* — an Active Reminder names a task or completion target that doesn't appear in =todo.org=, or =todo.org= names a reminder not present in *Active Reminders*.
+2. *Passed scheduled date* — a task with =SCHEDULED:= or =DEADLINE:= earlier than today is still =TODO= or =DOING= with no completion log under it.
+3. *"Waiting on X" matches a shipped X* — a task body contains "waiting on", "after X ships", "blocked by", and the recent session *Next Steps* (or *Files Modified*) summaries log the named X as shipped.
+4. *Dead =file:= link* — a task body's =file:= link points to a path that no longer exists (one existence check per linked path).
+5. *Sub-task DONE coverage >75%* — a parent task is still =TODO= or =DOING= while >75% of its sub-tasks are =DONE= or =CANCELLED= (a likely promotion-to-DONE candidate).
+
+** Surface and decide
+
+If neither trigger trips, proceed silently to Phase B.
+
+If one or both trip, surface them inline and ask:
+
+#+begin_example
+Task-audit looks worth running before recommending what's next:
+- Temporal: 23 days since last audit (threshold 14).
+- Signal: "blocked by review of spec X" — sessions show spec X merged.
+
+1. Run task-audit first, then come back to "what's next" (recommended)
+2. Skip the audit — proceed straight to the recommendation
+3. Defer — proceed now, file a TODO to run task-audit later
+#+end_example
+
+If the user picks option 1, hand off to =task-audit.org= and resume Next Mode at Phase B after it stamps =:LAST_AUDIT:=. If option 2, continue. If option 3, file a =** TODO [#B] Run task-audit= entry under =* $Project Open Work= with today's date and continue.
+
 * Phase B: Reconcile (both modes)
 
 Compare the open-task signal across sources. For each task in =notes.org= Active Reminders or in a recent session's Next Steps that does NOT have a corresponding =todo.org= entry:
@@ -118,61 +156,89 @@ For tasks Craig says are NOT done, leave them as-is.
 
 ** Next Mode
 
-Apply the prioritization cascade in order. Stop at the first matching step:
+Next Mode answers two questions in one output: "what matters most right now?" (the cascade, importance/urgency) and "what shape of task can I actually finish?" (the friction filter, effort + autonomy). The user picks the row that matches the state of their day.
+
+*** Step 1 — Cascade recommendation (importance/urgency)
 
-*** 1. In-Progress Tasks
+Apply the prioritization cascade in order. Stop at the first matching step. This is the importance/urgency answer.
+
+**** 1. In-Progress Tasks
 - Look for tasks marked =DOING= or partially complete.
 - *If found:* Recommend that task (always finish what's started).
 - *If user declines:* Continue to next step.
 
-*** 2. Active Reminders
+**** 2. Active Reminders
 - Review notes.org Active Reminders (already in the Phase A snapshot).
 - *If found:* Recommend reminder task.
 - *If user declines:* Add to =todo.org= per Phase B (if not already there), then continue.
 
-*** 3. Deadline-Driven Tasks
+**** 3. Deadline-Driven Tasks
 - Scan =todo.org= for tasks with explicit deadlines.
 - *If found:* Recommend the task with the closest deadline.
 - *If none:* Continue to next step.
 
-*** 4. V2MOM Method Order (if applicable)
+**** 4. V2MOM Method Order (if applicable)
 If =todo.org= is structured with V2MOM methods:
 - Method 1 priority A tasks first.
 - Then Method 2 priority A, Method 3 priority A, etc.
 - Then Method 1 priority B, Method 2 priority B, etc.
 - Continue pattern through priorities C and D.
 
-*** 5. Simple Priority Order
+**** 5. Simple Priority Order
 If =todo.org= is a flat list:
 - Evaluate all priority A tasks, pick most important.
 - If no priority A, evaluate priority B tasks.
 - Continue through priorities C and D.
 
-*** 6. All Tasks Complete
+**** 6. All Tasks Complete
 If no tasks remain: report "All done — no open tasks in =todo.org=."
 
-*** Handling Multiple Tasks at Same Level
+**** Handling Multiple Tasks at Same Level
 
 When multiple tasks share priority/method position, pick one based on:
 
 1. *Blocks other work* — dependencies matter.
 2. *Recently discussed* — mentioned in recent conversation.
 3. *Most foundational* — enables other tasks.
-4. *If truly uncertain* — show 2-3 options and let Craig choose.
+4. *If truly uncertain* — show 2-3 options and let the user choose.
+
+*** Step 2 — Friction filter (effort + autonomy)
+
+Independently of the cascade, scan the open-task set for tasks tagged =:quick:= and =:solo:=. Build three ranked picks:
+
+- *Quick + solo* — the top task carrying BOTH tags. Smallest effort × highest autonomy = lowest friction. Ideal for a 20-minute window or a flagging-energy moment.
+- *Quick only* — the top task carrying =:quick:= but not =:solo:=. Small but needs the user's preference / hardware / sign-off.
+- *Solo only* — the top task carrying =:solo:= but not =:quick:=. Bounded scope that Claude can finish end-to-end, but not a short job.
+
+Within each row, pick a single task per the same-level tie-breakers above (blocks-other-work, recently-discussed, most-foundational). If a row has no tasks, omit it rather than padding.
+
+The friction filter is the override path. When the cascade winner is partially blocked, hardware-dependent, or simply too large for the user's current state, one of the friction rows is what they pick instead.
 
 *** Output Format
 
-Keep the recommendation concise but informative:
+Pair the cascade recommendation with the friction block beneath it. Recommendation-at-item-1 convention applies to the friction rows — quick+solo first, since it's the strongest low-friction pick.
 
 #+begin_example
-Next: Fix org-noter reliability (Method 1, Priority A, 8/18 complete)
-Reason: Blocks daily reading/annotation workflow
+Cascade recommendation (importance/urgency):
+- Fix org-noter reliability — [#A], Method 1, 8/18 complete, blocks daily reading/annotation
+
+If you want lower friction instead:
+1. Quick + solo: Bump linter config — [#C] :quick:solo:, ~15 min
+2. Quick:       Confirm new dirvish setup — [#B] :quick:, needs your eye
+3. Solo:        Refactor config-utilities — [#B] :solo:, bounded but multi-hour
 #+end_example
 
-Include:
+Include for each row:
 - Task name / description.
-- One-line reasoning (which cascade step matched and why).
-- Progress indicator (for V2MOM-structured todos).
+- Priority + tag cluster.
+- One-line reasoning. For the cascade row, name which cascade step matched. For friction rows, an effort hint when one is obvious.
+- Progress indicator (for V2MOM-structured todos) on the cascade row only.
+
+**** Edge cases
+
+- *Empty friction block.* If no =:quick:= or =:solo:= tagged tasks exist in the open set, omit the friction block entirely. Present only the cascade recommendation.
+- *Dedupe.* If the cascade recommendation IS the same task as one of the friction rows (e.g. it's =:quick:solo:= and also won the cascade), show it once at the top with both labels. Don't list it twice.
+- *Decline behavior.* If the user declines the cascade recommendation, drop straight to the friction block as the natural next prompt. Do not fall through to lower-cascade-tier tasks; the friction filter IS the override.
 
 * Resolving a Task — Format Reference
 
@@ -204,8 +270,12 @@ Key elements:
 4. *Creating =todo.org= entries for pure informational reminders* — use judgment in Phase B.
 5. *Using a table for the task list* — Craig prefers flat bulleted lists in list mode.
 6. *Skipping the cascade order in next mode* — the cascade exists to override subjective choice with objective criteria; respect it.
-7. *Recommending more than one task in next mode* — be decisive. Only show 2-3 if truly uncertain after applying the same-level tie-breakers.
+7. *Recommending more than one task in the cascade row* — be decisive on the cascade. The friction block is a 3-row choice by design, but the cascade row is single-pick.
 8. *Re-querying =todo.org= during Phase C* — the snapshot from Phase A is canonical; don't re-read.
+9. *Skipping the friction block when the cascade winner isn't actionable.* The friction filter is the override path; don't fall through to lower-cascade-tier tasks if a =:quick:= or =:solo:= task is what's actually needed.
+10. *Padding the friction block when a row is empty.* If no task carries both =:quick:= and =:solo:=, omit that row entirely. Don't promote a =:quick:=-only task into the =:quick:solo:= slot.
+11. *Recommending over stale facts in Next Mode.* When =:LAST_AUDIT:= is old or a state signal trips in Phase A.1, offer =task-audit= first. Let the user override — but don't skip the offer.
+12. *Running Phase A.1 in List Mode.* The audit-warranted check is a Next Mode pre-step. List Mode displays the plate as-is and shouldn't trigger a write workflow first.
 
 * Living Document
 
diff --git a/.ai/workflows/task-audit.org b/.ai/workflows/task-audit.org
index edbaa8d..c999dab 100644
--- a/.ai/workflows/task-audit.org
+++ b/.ai/workflows/task-audit.org
@@ -73,7 +73,9 @@ For every STALE task, edit it in the main thread:
 - Mark statuses that moved; rewrite "waiting on X" lines whose X resolved.
 - Fix dead/renamed =file:= links.
 - *Consolidate duplicates* — when several tasks track the same thing, fold them into one home and delete the duplicates (per the user's call on which is canonical).
-- *Re-assess the =:quick:= and =:solo:= tags* — reconciliation can change a task's effort or autonomy: a resolved dependency may make a stuck task =:solo:=, a scope cut may make it =:quick:=, and new complexity surfaced by the sources can invalidate either. Add or remove the tags per the definitions in [[file:task-review.org][task-review.org]] when the reconciled facts make the call clear. When they don't — an effort estimate you can't pin down, a =:solo:= gate you can't confirm — it's a NEEDS-USER flag, not a guess.
+- *Ensure priority is set per the project scheme.* The top of the project's =todo.org= should carry the priority legend (=[#A]= through =[#D]=). Every task should carry an explicit priority cookie. If a cookie is missing, or no longer matches the reconciled facts, assign the right level per the legend. If the level is unambiguous from the body, do it autonomously; if it's a judgment call (especially the [#A] / [#B] line for important-but-not-urgent work), flag NEEDS-USER. Also enforce the [#A]-discipline rule from the legend — an [#A] task without a =SCHEDULED:= or =DEADLINE:= line is mis-graded and is either down-graded to [#B] (when reconciled facts say "important but not urgent") or surfaced as NEEDS-USER for the user to date.
+- *Ensure a type tag is set.* Every task carries one type tag from the project's tag legend (typically =:feature:= / =:chore:= / =:spec:= / =:bug:=). If missing or wrong, assign or correct it from the body when the type is unambiguous. If two tags fit (a refactor that also fixes a bug; a spec that's also a chore), flag NEEDS-USER rather than picking one silently.
+- *Re-assess the =:quick:= and =:solo:= tags* — reconciliation can change a task's effort or autonomy: a resolved dependency may make a stuck task =:solo:=, a scope cut may make it =:quick:=, and new complexity surfaced by the sources can invalidate either. Add or remove the tags per the definitions in the project's tag legend (and [[file:task-review.org][task-review.org]]) when the reconciled facts make the call clear. When they don't — an effort estimate you can't pin down, a =:solo:= gate you can't confirm — it's a NEEDS-USER flag, not a guess.
 - Bump =:LAST_REVIEWED:= on each edited task.
 
 Follow =todo-format.md= for completion mechanics (depth-based DONE vs dated-rewrite) and the working-files / link-hygiene rules when moving artifacts.
@@ -84,6 +86,18 @@ Present the NEEDS-USER bucket as a short, scannable list — one line per task,
 
 When the trail on a task has gone cold and only the user knows the outcome, *say so and ask* — do not invent a status. ("The screen was set for 5/15; nothing in sessions, email, or chat records what came of it. Did it happen, and what's your read?")
 
+** Phase E — Close out
+
+After Phase D's judgment calls are adjudicated and applied, stamp the audit run by updating =.ai/notes.org='s *Workflow State* section:
+
+#+begin_example
+:LAST_AUDIT: YYYY-MM-DD
+#+end_example
+
+Use today's =date= output (=date "+%Y-%m-%d"=). Overwrite the previous value — the marker reflects the most recent run, not a history.
+
+If the *Workflow State* section doesn't exist yet, create it (placement: after *Active Reminders*). Other workflows (notably =open-tasks.org= Next Mode's audit-warranted check) read this marker to decide whether to offer running another audit.
+
 * Common Mistakes
 
 1. *Guessing a judgment call instead of flagging it.* If the answer isn't in the sources, it's bucket 3. Ask.
diff --git a/claude-templates/.ai/workflows/open-tasks.org b/claude-templates/.ai/workflows/open-tasks.org
index 7143aea..e90bc12 100644
--- a/claude-templates/.ai/workflows/open-tasks.org
+++ b/claude-templates/.ai/workflows/open-tasks.org
@@ -25,13 +25,51 @@ Don't route "task review" / "review tasks" here — those trigger the hygiene ha
 
 Issue all source reads as a single batch of parallel tool calls. They're independent:
 
-1. Read =notes.org= — sections *Active Reminders* and *Pending Decisions* only.
+1. Read =notes.org= — sections *Active Reminders*, *Pending Decisions*, and *Workflow State* (note the =:LAST_AUDIT:= date if present; "unset" or absent is treated as "never run").
 2. Read each of the 2-3 most recent files in =.ai/sessions/= — only the *Next Steps* subsection under =* Summary= (those capture recent "next time, do X" items that may still be open).
 3. Read =todo.org= — entries under the open-work header (=* $Project Open Work=, e.g. =* Homelab Open Work=). Skip the resolved header (=* $Project Resolved=).
 4. Run =date= for an accurate timestamp (used to evaluate deadlines + flag suspected completions).
 
 The phases below work entirely from this in-memory snapshot. The original split-workflow versions issued these reads sequentially in different files; consolidating saves round-trips and prevents the two implementations drifting apart.
 
+* Phase A.1: Evaluate audit-warranted (Next Mode only)
+
+Before Next Mode produces a recommendation, decide whether a =task-audit= run looks warranted. List Mode skips this phase — it displays the plate as-is and should not trigger a write-side-effect workflow first.
+
+Two triggers; either one trips the warranted flag.
+
+** Temporal threshold
+
+If =:LAST_AUDIT:= is absent / unset, or its date is older than 14 days from today, the temporal trigger trips. (The 14-day default is per-project; tune by editing this workflow if a project churns faster or slower.)
+
+** State-signal scan
+
+Walk the Phase A snapshot for any of these cheap signals — one is enough:
+
+1. *Reminder/task mismatch* — an Active Reminder names a task or completion target that doesn't appear in =todo.org=, or =todo.org= names a reminder not present in *Active Reminders*.
+2. *Passed scheduled date* — a task with =SCHEDULED:= or =DEADLINE:= earlier than today is still =TODO= or =DOING= with no completion log under it.
+3. *"Waiting on X" matches a shipped X* — a task body contains "waiting on", "after X ships", "blocked by", and the recent session *Next Steps* (or *Files Modified*) summaries log the named X as shipped.
+4. *Dead =file:= link* — a task body's =file:= link points to a path that no longer exists (one existence check per linked path).
+5. *Sub-task DONE coverage >75%* — a parent task is still =TODO= or =DOING= while >75% of its sub-tasks are =DONE= or =CANCELLED= (a likely promotion-to-DONE candidate).
+
+** Surface and decide
+
+If neither trigger trips, proceed silently to Phase B.
+
+If one or both trip, surface them inline and ask:
+
+#+begin_example
+Task-audit looks worth running before recommending what's next:
+- Temporal: 23 days since last audit (threshold 14).
+- Signal: "blocked by review of spec X" — sessions show spec X merged.
+
+1. Run task-audit first, then come back to "what's next" (recommended)
+2. Skip the audit — proceed straight to the recommendation
+3. Defer — proceed now, file a TODO to run task-audit later
+#+end_example
+
+If the user picks option 1, hand off to =task-audit.org= and resume Next Mode at Phase B after it stamps =:LAST_AUDIT:=. If option 2, continue. If option 3, file a =** TODO [#B] Run task-audit= entry under =* $Project Open Work= with today's date and continue.
+
 * Phase B: Reconcile (both modes)
 
 Compare the open-task signal across sources. For each task in =notes.org= Active Reminders or in a recent session's Next Steps that does NOT have a corresponding =todo.org= entry:
@@ -236,6 +274,8 @@ Key elements:
 8. *Re-querying =todo.org= during Phase C* — the snapshot from Phase A is canonical; don't re-read.
 9. *Skipping the friction block when the cascade winner isn't actionable.* The friction filter is the override path; don't fall through to lower-cascade-tier tasks if a =:quick:= or =:solo:= task is what's actually needed.
 10. *Padding the friction block when a row is empty.* If no task carries both =:quick:= and =:solo:=, omit that row entirely. Don't promote a =:quick:=-only task into the =:quick:solo:= slot.
+11. *Recommending over stale facts in Next Mode.* When =:LAST_AUDIT:= is old or a state signal trips in Phase A.1, offer =task-audit= first. Let the user override — but don't skip the offer.
+12. *Running Phase A.1 in List Mode.* The audit-warranted check is a Next Mode pre-step. List Mode displays the plate as-is and shouldn't trigger a write workflow first.
 
 * Living Document
 
diff --git a/claude-templates/.ai/workflows/task-audit.org b/claude-templates/.ai/workflows/task-audit.org
index edbaa8d..c999dab 100644
--- a/claude-templates/.ai/workflows/task-audit.org
+++ b/claude-templates/.ai/workflows/task-audit.org
@@ -73,7 +73,9 @@ For every STALE task, edit it in the main thread:
 - Mark statuses that moved; rewrite "waiting on X" lines whose X resolved.
 - Fix dead/renamed =file:= links.
 - *Consolidate duplicates* — when several tasks track the same thing, fold them into one home and delete the duplicates (per the user's call on which is canonical).
-- *Re-assess the =:quick:= and =:solo:= tags* — reconciliation can change a task's effort or autonomy: a resolved dependency may make a stuck task =:solo:=, a scope cut may make it =:quick:=, and new complexity surfaced by the sources can invalidate either. Add or remove the tags per the definitions in [[file:task-review.org][task-review.org]] when the reconciled facts make the call clear. When they don't — an effort estimate you can't pin down, a =:solo:= gate you can't confirm — it's a NEEDS-USER flag, not a guess.
+- *Ensure priority is set per the project scheme.* The top of the project's =todo.org= should carry the priority legend (=[#A]= through =[#D]=). Every task should carry an explicit priority cookie. If a cookie is missing, or no longer matches the reconciled facts, assign the right level per the legend. If the level is unambiguous from the body, do it autonomously; if it's a judgment call (especially the [#A] / [#B] line for important-but-not-urgent work), flag NEEDS-USER. Also enforce the [#A]-discipline rule from the legend — an [#A] task without a =SCHEDULED:= or =DEADLINE:= line is mis-graded and is either down-graded to [#B] (when reconciled facts say "important but not urgent") or surfaced as NEEDS-USER for the user to date.
+- *Ensure a type tag is set.* Every task carries one type tag from the project's tag legend (typically =:feature:= / =:chore:= / =:spec:= / =:bug:=). If missing or wrong, assign or correct it from the body when the type is unambiguous. If two tags fit (a refactor that also fixes a bug; a spec that's also a chore), flag NEEDS-USER rather than picking one silently.
+- *Re-assess the =:quick:= and =:solo:= tags* — reconciliation can change a task's effort or autonomy: a resolved dependency may make a stuck task =:solo:=, a scope cut may make it =:quick:=, and new complexity surfaced by the sources can invalidate either. Add or remove the tags per the definitions in the project's tag legend (and [[file:task-review.org][task-review.org]]) when the reconciled facts make the call clear. When they don't — an effort estimate you can't pin down, a =:solo:= gate you can't confirm — it's a NEEDS-USER flag, not a guess.
 - Bump =:LAST_REVIEWED:= on each edited task.
 
 Follow =todo-format.md= for completion mechanics (depth-based DONE vs dated-rewrite) and the working-files / link-hygiene rules when moving artifacts.
@@ -84,6 +86,18 @@ Present the NEEDS-USER bucket as a short, scannable list — one line per task,
 
 When the trail on a task has gone cold and only the user knows the outcome, *say so and ask* — do not invent a status. ("The screen was set for 5/15; nothing in sessions, email, or chat records what came of it. Did it happen, and what's your read?")
 
+** Phase E — Close out
+
+After Phase D's judgment calls are adjudicated and applied, stamp the audit run by updating =.ai/notes.org='s *Workflow State* section:
+
+#+begin_example
+:LAST_AUDIT: YYYY-MM-DD
+#+end_example
+
+Use today's =date= output (=date "+%Y-%m-%d"=). Overwrite the previous value — the marker reflects the most recent run, not a history.
+
+If the *Workflow State* section doesn't exist yet, create it (placement: after *Active Reminders*). Other workflows (notably =open-tasks.org= Next Mode's audit-warranted check) read this marker to decide whether to offer running another audit.
+
 * Common Mistakes
 
 1. *Guessing a judgment call instead of flagging it.* If the answer isn't in the sources, it's bucket 3. Ask.