diff --git a/gardener/gardener-run.sh b/gardener/gardener-run.sh
index 31aa8c0..6c964dc 100755
--- a/gardener/gardener-run.sh
+++ b/gardener/gardener-run.sh
@@ -92,8 +92,11 @@ Supported actions:
 The commit-and-pr step converts JSONL to JSON array. The orchestrator executes
 actions after the PR merges. Do NOT call mutation APIs directly during the run."
 
-build_sdk_prompt_footer "$GARDENER_API_EXTRA"
-PROMPT_FOOTER="${PROMPT_FOOTER}## Completion protocol (REQUIRED)
+# Reuse shared footer (API reference + environment), replace phase protocol
+# shellcheck disable=SC2034  # consumed by build_prompt_footer
+PHASE_FILE=""  # not used in SDK mode
+build_prompt_footer "$GARDENER_API_EXTRA"
+PROMPT_FOOTER="${PROMPT_FOOTER%%## Phase protocol*}## Completion protocol (REQUIRED)
 When the commit-and-pr step creates a PR, write the PR number and stop:
   echo \"\$PR_NUMBER\" > '${GARDENER_PR_FILE}'
 Then STOP. Do NOT write PHASE: signals — the orchestrator handles CI, review, and merge.
diff --git a/lib/formula-session.sh b/lib/formula-session.sh
index 7c52035..670da95 100644
--- a/lib/formula-session.sh
+++ b/lib/formula-session.sh
@@ -291,33 +291,6 @@ build_graph_section() {
   fi
 }
 
-# ── SDK helpers ───────────────────────────────────────────────────────────
-
-# build_sdk_prompt_footer [EXTRA_API_LINES]
-# Like build_prompt_footer but omits the phase protocol section (SDK mode).
-# Sets PROMPT_FOOTER.
-build_sdk_prompt_footer() {
-  # shellcheck disable=SC2034  # consumed by build_prompt_footer
-  PHASE_FILE=""  # not used in SDK mode
-  build_prompt_footer "${1:-}"
-  PROMPT_FOOTER="${PROMPT_FOOTER%%## Phase protocol*}"
-}
-
-# formula_worktree_setup WORKTREE
-# Creates an isolated worktree for synchronous formula execution.
-# Fetches primary branch, cleans stale worktree, creates new one, and
-# sets an EXIT trap for cleanup.
-# Requires globals: PROJECT_REPO_ROOT, PRIMARY_BRANCH.
-formula_worktree_setup() {
-  local worktree="$1"
-  cd "$PROJECT_REPO_ROOT" || return
-  git fetch origin "$PRIMARY_BRANCH" 2>/dev/null || true
-  worktree_cleanup "$worktree"
-  git worktree add "$worktree" "origin/${PRIMARY_BRANCH}" --detach 2>/dev/null
-  # shellcheck disable=SC2064  # expand worktree now, not at trap time
-  trap "worktree_cleanup '$worktree'" EXIT
-}
-
 # ── Prompt + monitor helpers ──────────────────────────────────────────────
 
 # build_prompt_footer [EXTRA_API_LINES]
diff --git a/planner/planner-run.sh b/planner/planner-run.sh
index 313f6ef..878fdeb 100755
--- a/planner/planner-run.sh
+++ b/planner/planner-run.sh
@@ -1,16 +1,10 @@
 #!/usr/bin/env bash
 # =============================================================================
-# planner-run.sh — Cron wrapper: planner execution via SDK + formula
+# planner-run.sh — Cron wrapper: direct planner execution via Claude + formula
 #
-# Synchronous bash loop using claude -p (one-shot invocation).
-# No tmux sessions, no phase files — the bash script IS the state machine.
-#
-# Flow:
-#   1. Guards: cron lock, memory check
-#   2. Load formula (formulas/run-planner.toml)
-#   3. Context: VISION.md, AGENTS.md, ops:RESOURCES.md, structural graph,
-#      planner memory, journal entries
-#   4. agent_run(worktree, prompt) → Claude plans, may push knowledge updates
+# Runs daily (or on-demand). Guards against concurrent runs and low memory.
+# Creates a tmux session with Claude (opus) reading formulas/run-planner.toml.
+# No action issues — the planner is a nervous system component, not work.
 #
 # Usage:
 #   planner-run.sh [projects/disinto.toml]   # project config (default: disinto)
@@ -26,22 +20,24 @@ export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 source "$FACTORY_ROOT/lib/env.sh"
 # Use planner-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_PLANNER_TOKEN:-${FORGE_TOKEN}}"
+# shellcheck source=../lib/agent-session.sh
+source "$FACTORY_ROOT/lib/agent-session.sh"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
 source "$FACTORY_ROOT/lib/worktree.sh"
 # shellcheck source=../lib/guard.sh
 source "$FACTORY_ROOT/lib/guard.sh"
-# shellcheck source=../lib/agent-sdk.sh
-source "$FACTORY_ROOT/lib/agent-sdk.sh"
 
 LOG_FILE="$SCRIPT_DIR/planner.log"
-# shellcheck disable=SC2034  # consumed by agent-sdk.sh
-LOGFILE="$LOG_FILE"
-# shellcheck disable=SC2034  # consumed by agent-sdk.sh
-SID_FILE="/tmp/planner-session-${PROJECT_NAME}.sid"
+# shellcheck disable=SC2034  # consumed by run_formula_and_monitor
+SESSION_NAME="planner-${PROJECT_NAME}"
+PHASE_FILE="/tmp/planner-session-${PROJECT_NAME}.phase"
+
+# shellcheck disable=SC2034  # read by monitor_phase_loop in lib/agent-session.sh
+PHASE_POLL_INTERVAL=15
+
 SCRATCH_FILE="/tmp/planner-${PROJECT_NAME}-scratch.md"
-WORKTREE="/tmp/${PROJECT_NAME}-planner-run"
 
 log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
 
@@ -96,13 +92,14 @@ SCRATCH_CONTEXT=$(read_scratch_context "$SCRATCH_FILE")
 SCRATCH_INSTRUCTION=$(build_scratch_instruction "$SCRATCH_FILE")
 
 # ── Build prompt ─────────────────────────────────────────────────────────
-build_sdk_prompt_footer "
+build_prompt_footer "
   Relabel:     curl -sf -H \"Authorization: token \${FORGE_TOKEN}\" -X PUT -H 'Content-Type: application/json' '${FORGE_API}/issues/{number}/labels' -d '{\"labels\":[LABEL_ID]}'
   Comment:     curl -sf -H \"Authorization: token \${FORGE_TOKEN}\" -X POST -H 'Content-Type: application/json' '${FORGE_API}/issues/{number}/comments' -d '{\"body\":\"...\"}'
   Close:       curl -sf -H \"Authorization: token \${FORGE_TOKEN}\" -X PATCH -H 'Content-Type: application/json' '${FORGE_API}/issues/{number}' -d '{\"state\":\"closed\"}'
 "
 
-PROMPT="You are the strategic planner for ${FORGE_REPO}. Work through the formula below.
+# shellcheck disable=SC2034  # consumed by run_formula_and_monitor
+PROMPT="You are the strategic planner for ${FORGE_REPO}. Work through the formula below. You MUST write PHASE:done to '${PHASE_FILE}' when finished — the orchestrator will time you out if you return to the prompt without signalling.
 
 ## Project context
 ${CONTEXT_BLOCK}${MEMORY_BLOCK}${JOURNAL_BLOCK}
@@ -116,14 +113,12 @@ ${SCRATCH_INSTRUCTION}
 
 ${PROMPT_FOOTER}"
 
-# ── Create worktree ──────────────────────────────────────────────────────
-formula_worktree_setup "$WORKTREE"
-
-# ── Run agent ─────────────────────────────────────────────────────────────
+# ── Run session ──────────────────────────────────────────────────────────
 export CLAUDE_MODEL="opus"
+run_formula_and_monitor "planner"
 
-agent_run --worktree "$WORKTREE" "$PROMPT"
-log "agent_run complete"
-
-rm -f "$SCRATCH_FILE"
-log "--- Planner run done ---"
+# ── Cleanup scratch file on normal exit ──────────────────────────────────
+# FINAL_PHASE already set by run_formula_and_monitor
+if [ "${FINAL_PHASE:-}" = "PHASE:done" ]; then
+  rm -f "$SCRATCH_FILE"
+fi
diff --git a/predictor/predictor-run.sh b/predictor/predictor-run.sh
index fb9bf51..c7921c0 100755
--- a/predictor/predictor-run.sh
+++ b/predictor/predictor-run.sh
@@ -1,15 +1,10 @@
 #!/usr/bin/env bash
 # =============================================================================
-# predictor-run.sh — Cron wrapper: predictor execution via SDK + formula
+# predictor-run.sh — Cron wrapper: predictor execution via Claude + formula
 #
-# Synchronous bash loop using claude -p (one-shot invocation).
-# No tmux sessions, no phase files — the bash script IS the state machine.
-#
-# Flow:
-#   1. Guards: cron lock, memory check
-#   2. Load formula (formulas/run-predictor.toml)
-#   3. Context: AGENTS.md, ops:RESOURCES.md, VISION.md, structural graph
-#   4. agent_run(worktree, prompt) → Claude analyzes, writes to ops repo
+# Runs daily (or on-demand). Guards against concurrent runs and low memory.
+# Creates a tmux session with Claude (sonnet) reading formulas/run-predictor.toml.
+# Files prediction/unreviewed issues for the planner to triage.
 #
 # Usage:
 #   predictor-run.sh [projects/disinto.toml]   # project config (default: disinto)
@@ -27,22 +22,24 @@ export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 source "$FACTORY_ROOT/lib/env.sh"
 # Use predictor-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_PREDICTOR_TOKEN:-${FORGE_TOKEN}}"
+# shellcheck source=../lib/agent-session.sh
+source "$FACTORY_ROOT/lib/agent-session.sh"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
 source "$FACTORY_ROOT/lib/worktree.sh"
 # shellcheck source=../lib/guard.sh
 source "$FACTORY_ROOT/lib/guard.sh"
-# shellcheck source=../lib/agent-sdk.sh
-source "$FACTORY_ROOT/lib/agent-sdk.sh"
 
 LOG_FILE="$SCRIPT_DIR/predictor.log"
-# shellcheck disable=SC2034  # consumed by agent-sdk.sh
-LOGFILE="$LOG_FILE"
-# shellcheck disable=SC2034  # consumed by agent-sdk.sh
-SID_FILE="/tmp/predictor-session-${PROJECT_NAME}.sid"
+# shellcheck disable=SC2034  # consumed by run_formula_and_monitor
+SESSION_NAME="predictor-${PROJECT_NAME}"
+PHASE_FILE="/tmp/predictor-session-${PROJECT_NAME}.phase"
+
+# shellcheck disable=SC2034  # read by monitor_phase_loop in lib/agent-session.sh
+PHASE_POLL_INTERVAL=15
+
 SCRATCH_FILE="/tmp/predictor-${PROJECT_NAME}-scratch.md"
-WORKTREE="/tmp/${PROJECT_NAME}-predictor-run"
 
 log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
 
@@ -65,10 +62,10 @@ SCRATCH_CONTEXT=$(read_scratch_context "$SCRATCH_FILE")
 SCRATCH_INSTRUCTION=$(build_scratch_instruction "$SCRATCH_FILE")
 
 # ── Build prompt ─────────────────────────────────────────────────────────
-build_sdk_prompt_footer
-export CLAUDE_MODEL="sonnet"
+build_prompt_footer
 
-PROMPT="You are the prediction agent (goblin) for ${FORGE_REPO}. Work through the formula below.
+# shellcheck disable=SC2034  # consumed by run_formula_and_monitor
+PROMPT="You are the prediction agent (goblin) for ${FORGE_REPO}. Work through the formula below. You MUST write PHASE:done to '${PHASE_FILE}' when finished — the orchestrator will time you out if you return to the prompt without signalling.
 
 Your role: abstract adversary. Find the project's biggest weakness, challenge
 planner claims, and generate evidence. Explore when uncertain (file a prediction),
@@ -91,12 +88,12 @@ ${FORMULA_CONTENT}
 ${SCRATCH_INSTRUCTION}
 ${PROMPT_FOOTER}"
 
-# ── Create worktree ──────────────────────────────────────────────────────
-formula_worktree_setup "$WORKTREE"
+# ── Run session ──────────────────────────────────────────────────────────
+export CLAUDE_MODEL="sonnet"
+run_formula_and_monitor "predictor"
 
-# ── Run agent ─────────────────────────────────────────────────────────────
-agent_run --worktree "$WORKTREE" "$PROMPT"
-log "agent_run complete"
-
-rm -f "$SCRATCH_FILE"
-log "--- Predictor run done ---"
+# ── Cleanup scratch file on normal exit ──────────────────────────────────
+# FINAL_PHASE already set by run_formula_and_monitor
+if [ "${FINAL_PHASE:-}" = "PHASE:done" ]; then
+  rm -f "$SCRATCH_FILE"
+fi
diff --git a/supervisor/supervisor-run.sh b/supervisor/supervisor-run.sh
index 129666f..d32bd79 100755
--- a/supervisor/supervisor-run.sh
+++ b/supervisor/supervisor-run.sh
@@ -1,17 +1,14 @@
 #!/usr/bin/env bash
 # =============================================================================
-# supervisor-run.sh — Cron wrapper: supervisor execution via SDK + formula
+# supervisor-run.sh — Cron wrapper: supervisor execution via Claude + formula
 #
-# Synchronous bash loop using claude -p (one-shot invocation).
-# No tmux sessions, no phase files — the bash script IS the state machine.
+# Runs every 20 minutes (or on-demand). Guards against concurrent runs and
+# low memory. Collects metrics via preflight.sh, then creates a tmux session
+# with Claude (sonnet) reading formulas/run-supervisor.toml.
 #
-# Flow:
-#   1. Guards: cron lock, memory check
-#   2. Housekeeping: clean up stale crashed worktrees
-#   3. Collect pre-flight metrics (supervisor/preflight.sh)
-#   4. Load formula (formulas/run-supervisor.toml)
-#   5. Context: AGENTS.md, preflight metrics, structural graph
-#   6. agent_run(worktree, prompt) → Claude monitors, may clean up
+# Replaces supervisor-poll.sh (bash orchestrator + claude -p one-shot) with
+# formula-driven interactive Claude session matching the planner/predictor
+# pattern.
 #
 # Usage:
 #   supervisor-run.sh [projects/disinto.toml]   # project config (default: disinto)
@@ -29,22 +26,24 @@ export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 source "$FACTORY_ROOT/lib/env.sh"
 # Use supervisor-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_SUPERVISOR_TOKEN:-${FORGE_TOKEN}}"
+# shellcheck source=../lib/agent-session.sh
+source "$FACTORY_ROOT/lib/agent-session.sh"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
 source "$FACTORY_ROOT/lib/worktree.sh"
 # shellcheck source=../lib/guard.sh
 source "$FACTORY_ROOT/lib/guard.sh"
-# shellcheck source=../lib/agent-sdk.sh
-source "$FACTORY_ROOT/lib/agent-sdk.sh"
 
 LOG_FILE="$SCRIPT_DIR/supervisor.log"
-# shellcheck disable=SC2034  # consumed by agent-sdk.sh
-LOGFILE="$LOG_FILE"
-# shellcheck disable=SC2034  # consumed by agent-sdk.sh
-SID_FILE="/tmp/supervisor-session-${PROJECT_NAME}.sid"
+# shellcheck disable=SC2034  # consumed by run_formula_and_monitor
+SESSION_NAME="supervisor-${PROJECT_NAME}"
+PHASE_FILE="/tmp/supervisor-session-${PROJECT_NAME}.phase"
+
+# shellcheck disable=SC2034  # read by monitor_phase_loop in lib/agent-session.sh
+PHASE_POLL_INTERVAL=15
+
 SCRATCH_FILE="/tmp/supervisor-${PROJECT_NAME}-scratch.md"
-WORKTREE="/tmp/${PROJECT_NAME}-supervisor-run"
 
 log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
 
@@ -76,13 +75,10 @@ SCRATCH_CONTEXT=$(read_scratch_context "$SCRATCH_FILE")
 SCRATCH_INSTRUCTION=$(build_scratch_instruction "$SCRATCH_FILE")
 
 # ── Build prompt ─────────────────────────────────────────────────────────
-build_sdk_prompt_footer
-export CLAUDE_MODEL="sonnet"
+build_prompt_footer
 
-# ── Create worktree (before prompt assembly so trap is set early) ────────
-formula_worktree_setup "$WORKTREE"
-
-PROMPT="You are the supervisor agent for ${FORGE_REPO}. Work through the formula below.
+# shellcheck disable=SC2034  # consumed by run_formula_and_monitor
+PROMPT="You are the supervisor agent for ${FORGE_REPO}. Work through the formula below. You MUST write PHASE:done to '${PHASE_FILE}' when finished — the orchestrator will time you out if you return to the prompt without signalling.
 
 You have full shell access and --dangerously-skip-permissions.
 Fix what you can. File vault items for what you cannot. Do NOT ask permission — act first, report after.
@@ -101,9 +97,12 @@ ${FORMULA_CONTENT}
 ${SCRATCH_INSTRUCTION}
 ${PROMPT_FOOTER}"
 
-# ── Run agent ─────────────────────────────────────────────────────────────
-agent_run --worktree "$WORKTREE" "$PROMPT"
-log "agent_run complete"
+# ── Run session ──────────────────────────────────────────────────────────
+export CLAUDE_MODEL="sonnet"
+run_formula_and_monitor "supervisor" 1200
 
-rm -f "$SCRATCH_FILE"
-log "--- Supervisor run done ---"
+# ── Cleanup scratch file on normal exit ──────────────────────────────────
+# FINAL_PHASE already set by run_formula_and_monitor
+if [ "${FINAL_PHASE:-}" = "PHASE:done" ]; then
+  rm -f "$SCRATCH_FILE"
+fi