fix: gardener migration — run-gardener.toml via direct cron, remove legacy scripts (#490)

Rewrite gardener-run.sh as direct cron runner (matching supervisor/planner/
predictor pattern): lock guard, memory check, worktree, tmux session with
Claude sonnet + formulas/run-gardener.toml, phase monitoring, cleanup.

- Delete gardener-poll.sh and gardener-agent.sh (superseded)
- Extract consume_escalation_reply() to lib/formula-session.sh (shared
  by gardener and supervisor, eliminates duplicate blocks)
- Update AGENTS.md, gardener/AGENTS.md, lib/AGENTS.md, CI smoke test,
  and cross-references

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
openhands 2026-03-21 13:09:17 +00:00
parent 685ca1034a
commit b630c6fcc1
11 changed files with 125 additions and 581 deletions

View file

@ -192,8 +192,6 @@ check_script dev/dev-agent.sh dev/phase-handler.sh
check_script dev/phase-handler.sh dev/dev-agent.sh lib/secret-scan.sh
check_script dev/dev-poll.sh
check_script dev/phase-test.sh
check_script gardener/gardener-agent.sh lib/agent-session.sh
check_script gardener/gardener-poll.sh
check_script gardener/gardener-run.sh
check_script review/review-pr.sh
check_script review/review-poll.sh

View file

@ -16,8 +16,7 @@ See `README.md` for the full architecture and `BOOTSTRAP.md` for setup.
disinto/
├── dev/ dev-poll.sh, dev-agent.sh, phase-handler.sh — issue implementation
├── review/ review-poll.sh, review-pr.sh — PR review
├── gardener/ gardener-run.sh — files action issue for run-gardener formula
│ gardener-poll.sh, gardener-agent.sh — grooming
├── gardener/ gardener-run.sh — direct cron executor for run-gardener formula
├── predictor/ predictor-run.sh — daily cron executor for run-predictor formula
├── planner/ planner-run.sh — direct cron executor for run-planner formula
│ planner/journal/ — daily raw logs from each planner run

View file

@ -627,7 +627,7 @@ Instructions:
"⚠️ Issue #${ISSUE} (PR #${PR_NUMBER:-none}) needs human input.${HUMAN_REASON:+ Reason: ${HUMAN_REASON}}" \
"⚠️ <a href='${_issue_url}'>Issue #${ISSUE}</a>${_pr_link} needs human input.${HUMAN_REASON:+ Reason: ${HUMAN_REASON}}<br>Reply in this thread to send guidance to the dev agent."
log "phase: needs_human — notified via Matrix, waiting for external injection"
# Don't inject anything — supervisor-poll.sh (#81) injects human replies, gardener-poll.sh as backup
# Don't inject anything — supervisor-run.sh (#81) injects human replies
# ── PHASE: done ─────────────────────────────────────────────────────────────
# PR merged and issue closed (by orchestrator or Claude). Just clean up local state.

View file

@ -1,7 +1,7 @@
# formulas/run-gardener.toml — Gardener housekeeping formula
#
# Defines the gardener's complete run: grooming (Claude session via
# gardener-agent.sh) + blocked-review + AGENTS.md maintenance + final
# gardener-run.sh) + blocked-review + AGENTS.md maintenance + final
# commit-and-pr.
#
# No memory, no journal. The gardener does mechanical housekeeping
@ -48,7 +48,7 @@ id = "grooming"
title = "Backlog grooming — triage all open issues"
description = """
Groom the open issue backlog. This step is the core Claude-driven analysis
(currently implemented in gardener-agent.sh with bash pre-checks).
(Claude performs pre-checks inline before deeper analysis).
Pre-checks (bash, zero tokens detect problems before invoking Claude):

View file

@ -5,19 +5,25 @@
criteria, oversized issues, stale issues, and circular dependencies. Invoke
Claude to fix or escalate to a human via Matrix.
**Trigger**: `gardener-run.sh` runs 2x/day via cron. It files an `action`
issue referencing `formulas/run-gardener.toml`; the action-agent picks it up
and executes the gardener steps in an interactive Claude tmux session. Accepts
an optional project TOML argument (configures which project the action issue is
filed against).
**Trigger**: `gardener-run.sh` runs 4x/day via cron. It creates a tmux
session with `claude --model sonnet`, injects `formulas/run-gardener.toml`
with escalation replies as context, monitors the phase file, and cleans up
on completion or timeout (2h max session). No action issues — the gardener
runs directly from cron like the planner, predictor, and supervisor.
**Key files**:
- `gardener/gardener-run.sh` — Cron wrapper: lock, memory guard, dedup check, files action issue
- `gardener/gardener-poll.sh` — Escalation-reply injection for dev sessions, invokes gardener-agent.sh for grooming
- `gardener/gardener-agent.sh` — Orchestrator: bash pre-analysis, creates tmux session (`gardener-{project}`) with interactive `claude`, monitors phase file, parses result file (ACTION:/DUST:/ESCALATE)
- `gardener/gardener-run.sh` — Cron wrapper + orchestrator: lock, memory guard,
consumes escalation replies, sources disinto project config, creates tmux session,
injects formula prompt, monitors phase file, handles crash recovery via
`run_formula_and_monitor`
- `formulas/run-gardener.toml` — Execution spec: preflight, grooming, dust-bundling, blocked-review, agents-update, commit-and-pr
**Environment variables consumed**:
- `CODEBERG_TOKEN`, `CODEBERG_REPO`, `CODEBERG_API`, `PROJECT_NAME`, `PROJECT_REPO_ROOT`
- `CLAUDE_TIMEOUT`
- `PRIMARY_BRANCH`, `CLAUDE_MODEL` (set to sonnet by gardener-run.sh)
- `MATRIX_TOKEN`, `MATRIX_ROOM_ID`, `MATRIX_HOMESERVER`
**Lifecycle**: gardener-run.sh (cron 0,6,12,18) → lock + memory guard →
consume escalation replies → load formula + context → create tmux session →
Claude grooms backlog, bundles dust, reviews blocked issues, updates AGENTS.md,
commits and creates PR → `PHASE:done`.

View file

@ -1,359 +0,0 @@
#!/usr/bin/env bash
# gardener-agent.sh — tmux + Claude interactive gardener session manager
#
# Usage: ./gardener-agent.sh [project-toml]
# Called by: gardener-poll.sh
#
# Lifecycle:
# 1. Read escalation reply (from ESCALATION_REPLY env var)
# 2. Fetch open issues + bash pre-checks (zero tokens)
# 3. If no problems detected, exit 0
# 4. Build prompt with result-file output + phase protocol instructions
# 5. Create tmux session: gardener-{project} with interactive claude
# 6. Inject prompt via tmux
# 7. Monitor phase file — Claude writes PHASE:done when finished
# 8. Parse result file (ACTION:/DUST:/ESCALATE) → Matrix + dust.jsonl
# 9. Dust bundling: groups with 3+ items → one backlog issue
#
# Phase file: /tmp/gardener-session-{project}.phase
# Result file: /tmp/gardener-result-{project}.txt
# Session: gardener-{project} (tmux)
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
export PROJECT_TOML="${1:-}"
# shellcheck source=../lib/env.sh
source "$FACTORY_ROOT/lib/env.sh"
# shellcheck source=../lib/agent-session.sh
source "$FACTORY_ROOT/lib/agent-session.sh"
# shellcheck source=../lib/formula-session.sh
source "$FACTORY_ROOT/lib/formula-session.sh"
LOG_FILE="$SCRIPT_DIR/gardener.log"
SESSION_NAME="gardener-${PROJECT_NAME}"
PHASE_FILE="/tmp/gardener-session-${PROJECT_NAME}.phase"
RESULT_FILE="/tmp/gardener-result-${PROJECT_NAME}.txt"
SCRATCH_FILE="/tmp/gardener-${PROJECT_NAME}-scratch.md"
# shellcheck disable=SC2034 # read by monitor_phase_loop in lib/agent-session.sh
PHASE_POLL_INTERVAL=15
log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
# Gitea labels API requires []int64 — look up the "backlog" label ID once
# Falls back to the known Codeberg repo ID if the API call fails
BACKLOG_LABEL_ID=$(codeberg_api GET "/labels" 2>/dev/null \
| jq -r '.[] | select(.name == "backlog") | .id' 2>/dev/null || true)
BACKLOG_LABEL_ID="${BACKLOG_LABEL_ID:-1300815}"
log "--- gardener-agent start ---"
# ── Read escalation reply (passed via env by gardener-poll.sh) ────────────
ESCALATION_REPLY="${ESCALATION_REPLY:-}"
# ── Fetch all open issues ─────────────────────────────────────────────────
ISSUES_JSON=$(codeberg_api GET "/issues?state=open&type=issues&limit=50&sort=updated&direction=desc" 2>/dev/null || true)
if [ -z "$ISSUES_JSON" ] || [ "$ISSUES_JSON" = "null" ]; then
log "Failed to fetch issues"
exit 1
fi
ISSUE_COUNT=$(echo "$ISSUES_JSON" | jq 'length')
log "Found $ISSUE_COUNT open issues"
if [ "$ISSUE_COUNT" -eq 0 ]; then
log "No open issues — nothing to groom"
exit 0
fi
# ── Bash pre-checks (zero tokens) ────────────────────────────────────────
PROBLEMS=""
# 1. Duplicate detection: issues with very similar titles
TITLES=$(echo "$ISSUES_JSON" | jq -r '.[] | "\(.number)\t\(.title)"')
DUPES=""
while IFS=$'\t' read -r num1 title1; do
while IFS=$'\t' read -r num2 title2; do
[ "$num1" -ge "$num2" ] && continue
# Normalize: lowercase, strip prefixes + series names, collapse whitespace
t1=$(echo "$title1" | tr '[:upper:]' '[:lower:]' | sed 's/^feat:\|^fix:\|^refactor://;s/llm seed[^—]*—\s*//;s/push3 evolution[^—]*—\s*//;s/[^a-z0-9 ]//g;s/ */ /g')
t2=$(echo "$title2" | tr '[:upper:]' '[:lower:]' | sed 's/^feat:\|^fix:\|^refactor://;s/llm seed[^—]*—\s*//;s/push3 evolution[^—]*—\s*//;s/[^a-z0-9 ]//g;s/ */ /g')
# Count shared words (>60% overlap = suspect)
WORDS1=$(echo "$t1" | tr ' ' '\n' | sort -u)
WORDS2=$(echo "$t2" | tr ' ' '\n' | sort -u)
SHARED=$(comm -12 <(echo "$WORDS1") <(echo "$WORDS2") | wc -l)
TOTAL1=$(echo "$WORDS1" | wc -l)
TOTAL2=$(echo "$WORDS2" | wc -l)
MIN_TOTAL=$(( TOTAL1 < TOTAL2 ? TOTAL1 : TOTAL2 ))
if [ "$MIN_TOTAL" -gt 2 ] && [ "$SHARED" -gt 0 ]; then
OVERLAP=$(( SHARED * 100 / MIN_TOTAL ))
if [ "$OVERLAP" -ge 60 ]; then
DUPES="${DUPES}possible_dupe: #${num1} vs #${num2} (${OVERLAP}% word overlap)\n"
fi
fi
done <<< "$TITLES"
done <<< "$TITLES"
[ -n "$DUPES" ] && PROBLEMS="${PROBLEMS}${DUPES}"
# 2. Missing acceptance criteria: issues with short body and no checkboxes
while IFS=$'\t' read -r num body_len has_checkbox; do
if [ "$body_len" -lt 100 ] && [ "$has_checkbox" = "false" ]; then
PROBLEMS="${PROBLEMS}thin_issue: #${num} — body < 100 chars, no acceptance criteria\n"
fi
done < <(echo "$ISSUES_JSON" | jq -r '.[] | "\(.number)\t\(.body | length)\t\(.body | test("- \\[[ x]\\]") // false)"')
# 3. Stale issues: no update in 14+ days
NOW_EPOCH=$(date +%s)
while IFS=$'\t' read -r num updated_at; do
UPDATED_EPOCH=$(date -d "$updated_at" +%s 2>/dev/null || echo 0)
AGE_DAYS=$(( (NOW_EPOCH - UPDATED_EPOCH) / 86400 ))
if [ "$AGE_DAYS" -ge 14 ]; then
PROBLEMS="${PROBLEMS}stale: #${num} — no activity for ${AGE_DAYS} days\n"
fi
done < <(echo "$ISSUES_JSON" | jq -r '.[] | "\(.number)\t\(.updated_at)"')
# 5. Blocker detection: find issues blocking backlog items that aren't themselves backlog
# This is the HIGHEST PRIORITY — a non-backlog blocker starves the entire factory
BACKLOG_ISSUES=$(echo "$ISSUES_JSON" | jq -r '.[] | select(.labels | map(.name) | index("backlog")) | .number')
BLOCKER_NUMS=""
for BNUM in $BACKLOG_ISSUES; do
BBODY=$(echo "$ISSUES_JSON" | jq -r --arg n "$BNUM" '.[] | select(.number == ($n | tonumber)) | .body // ""')
# Extract deps from ## Dependencies / ## Depends on / ## Blocked by
IN_SECTION=false
while IFS= read -r line; do
if echo "$line" | grep -qiP '^##?\s*(Dependencies|Depends on|Blocked by)'; then IN_SECTION=true; continue; fi
if echo "$line" | grep -qP '^##?\s' && [ "$IN_SECTION" = true ]; then IN_SECTION=false; fi
if [ "$IN_SECTION" = true ]; then
for dep in $(echo "$line" | grep -oP '#\d+' | grep -oP '\d+'); do
[ "$dep" = "$BNUM" ] && continue
# Check if dep is open but NOT backlog-labeled
DEP_STATE=$(echo "$ISSUES_JSON" | jq -r --arg n "$dep" '.[] | select(.number == ($n | tonumber)) | .state' 2>/dev/null || true)
DEP_LABELS=$(echo "$ISSUES_JSON" | jq -r --arg n "$dep" '.[] | select(.number == ($n | tonumber)) | [.labels[].name] | join(",")' 2>/dev/null || true)
if [ "$DEP_STATE" = "open" ] && ! echo ",$DEP_LABELS," | grep -q ',backlog,'; then
BLOCKER_NUMS="${BLOCKER_NUMS} ${dep}"
fi
done
fi
done <<< "$BBODY"
done
# Deduplicate blockers
BLOCKER_NUMS=$(echo "$BLOCKER_NUMS" | tr ' ' '\n' | sort -un | head -10)
if [ -n "$BLOCKER_NUMS" ]; then
BLOCKER_LIST=""
for bnum in $BLOCKER_NUMS; do
BTITLE=$(echo "$ISSUES_JSON" | jq -r --arg n "$bnum" '.[] | select(.number == ($n | tonumber)) | .title' 2>/dev/null || true)
BLABELS=$(echo "$ISSUES_JSON" | jq -r --arg n "$bnum" '.[] | select(.number == ($n | tonumber)) | [.labels[].name] | join(",")' 2>/dev/null || true)
BLOCKER_LIST="${BLOCKER_LIST}#${bnum} [${BLABELS:-unlabeled}] ${BTITLE}\n"
done
PROBLEMS="${PROBLEMS}PRIORITY_blockers_starving_factory: these issues block backlog items but are NOT labeled backlog — promote them FIRST:\n${BLOCKER_LIST}\n"
fi
# 6. Tech-debt issues needing promotion to backlog (secondary to blockers)
TECH_DEBT_ISSUES=$(echo "$ISSUES_JSON" | jq -r '.[] | select(.labels | map(.name) | index("tech-debt")) | "#\(.number) \(.title)"')
if [ -n "$TECH_DEBT_ISSUES" ]; then
TECH_DEBT_COUNT=$(echo "$TECH_DEBT_ISSUES" | wc -l)
PROBLEMS="${PROBLEMS}tech_debt_promotion: ${TECH_DEBT_COUNT} tech-debt issues need processing (goal: zero tech-debt):\n$(echo "$TECH_DEBT_ISSUES" | head -50)\n"
fi
PROBLEM_COUNT=$(echo -e "$PROBLEMS" | grep -c '.' || true)
log "Detected $PROBLEM_COUNT potential problems"
if [ "$PROBLEM_COUNT" -eq 0 ] && [ -z "$ESCALATION_REPLY" ]; then
log "Backlog is clean — nothing to groom"
exit 0
fi
# ── Load formula ─────────────────────────────────────────────────────────
log "Loading groom-backlog formula"
FORMULA_FILE="$FACTORY_ROOT/formulas/groom-backlog.toml"
if [ ! -f "$FORMULA_FILE" ]; then
log "ERROR: formula not found: $FORMULA_FILE"
exit 1
fi
FORMULA_CONTENT=$(cat "$FORMULA_FILE")
# ── Read context files from project root ──────────────────────────────────
CONTEXT_BLOCK=""
for ctx in README.md AGENTS.md VISION.md; do
ctx_path="${PROJECT_REPO_ROOT}/${ctx}"
if [ -f "$ctx_path" ]; then
CONTEXT_BLOCK="${CONTEXT_BLOCK}
### ${ctx}
$(cat "$ctx_path")
"
fi
done
# ── Build issue context ────────────────────────────────────────────────────
ISSUE_SUMMARY=$(echo "$ISSUES_JSON" | jq -r '.[] | "#\(.number) [\(.labels | map(.name) | join(","))] \(.title)"')
# ── Build optional prompt sections ────────────────────────────────────────
CONTEXT_SECTION=""
if [ -n "$CONTEXT_BLOCK" ]; then
CONTEXT_SECTION="## Project context
${CONTEXT_BLOCK}"
fi
ESCALATION_SECTION=""
if [ -n "$ESCALATION_REPLY" ]; then
ESCALATION_SECTION="
### Human response to previous escalation
Format: '1a 2c 3b' means question 1→option (a), 2→option (c), 3→option (b).
Execute each chosen option via the Codeberg API FIRST, before processing new items.
If a choice is unclear, re-escalate that single item with a clarifying question.
${ESCALATION_REPLY}"
fi
# ── Read scratch file (compaction survival) ───────────────────────────────
SCRATCH_CONTEXT=$(read_scratch_context "$SCRATCH_FILE")
SCRATCH_INSTRUCTION=$(build_scratch_instruction "$SCRATCH_FILE")
# ── Build prompt from formula + dynamic context ────────────────────────────
log "Building gardener prompt from formula"
PROMPT="You are the issue gardener for ${CODEBERG_REPO}. Work through the formula below. You MUST write PHASE:done to '${PHASE_FILE}' when finished — the orchestrator will time you out if you return to the prompt without signalling.
${CONTEXT_SECTION}
${SCRATCH_CONTEXT:+${SCRATCH_CONTEXT}
}## Formula
${FORMULA_CONTENT}
## Runtime context (bash pre-analysis)
### All open issues
${ISSUE_SUMMARY}
### Problems detected
$(echo -e "$PROBLEMS")${ESCALATION_SECTION}
## Codeberg API reference
Base URL: ${CODEBERG_API}
Auth header: -H \"Authorization: token \$CODEBERG_TOKEN\"
Read issue: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" '${CODEBERG_API}/issues/{number}' | jq '.body'
Relabel: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X PUT -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}/labels' -d '{\"labels\":[LABEL_ID]}'
Comment: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X POST -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}/comments' -d '{\"body\":\"...\"}'
Close: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X PATCH -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}' -d '{\"state\":\"closed\"}'
Edit body: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X PATCH -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}' -d '{\"body\":\"new body\"}'
List labels: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" '${CODEBERG_API}/labels'
NEVER echo or include the actual token value in output — always reference \$CODEBERG_TOKEN.
## Output format (MANDATORY — write each line to result file using bash)
echo \"ACTION: description of what you did\" >> '${RESULT_FILE}'
echo 'DUST: {\"issue\": NNN, \"group\": \"...\", \"title\": \"...\", \"reason\": \"...\"}' >> '${RESULT_FILE}'
printf 'ESCALATE\n1. #NNN \"title\" — reason (a) option1 (b) option2\n' >> '${RESULT_FILE}'
echo 'CLEAN' >> '${RESULT_FILE}' # only if truly nothing to do
${SCRATCH_INSTRUCTION}
## Phase protocol (REQUIRED)
When all work is done and verify confirms zero tech-debt:
echo 'PHASE:done' > '${PHASE_FILE}'
On unrecoverable error:
printf 'PHASE:failed\nReason: %s\n' 'describe error' > '${PHASE_FILE}'"
# ── Reset phase + result files ────────────────────────────────────────────
agent_kill_session "$SESSION_NAME"
rm -f "$PHASE_FILE" "${PHASE_FILE%.phase}.context" "$RESULT_FILE"
touch "$RESULT_FILE"
# ── Create tmux session ───────────────────────────────────────────────────
log "Creating tmux session: ${SESSION_NAME}"
if ! create_agent_session "$SESSION_NAME" "$PROJECT_REPO_ROOT" "$PHASE_FILE"; then
log "ERROR: failed to create tmux session ${SESSION_NAME}"
exit 1
fi
# Write phase protocol to context file for compaction survival
write_compact_context "$PHASE_FILE" "## Phase protocol (REQUIRED)
When all work is done and verify confirms zero tech-debt:
echo 'PHASE:done' > '${PHASE_FILE}'
On unrecoverable error:
printf 'PHASE:failed\nReason: %s\n' 'describe error' > '${PHASE_FILE}'"
agent_inject_into_session "$SESSION_NAME" "$PROMPT"
log "Prompt sent to tmux session"
matrix_send "gardener" "🌱 Gardener session started for ${CODEBERG_REPO}" 2>/dev/null || true
# ── Phase monitoring loop ─────────────────────────────────────────────────
log "Monitoring phase file: ${PHASE_FILE}"
_FORMULA_CRASH_COUNT=0
gardener_phase_callback() {
# Gardener-specific cleanup before shared crash recovery
if [ "$1" = "PHASE:crashed" ]; then
rm -f "$RESULT_FILE"
touch "$RESULT_FILE"
fi
formula_phase_callback "$1"
}
monitor_phase_loop "$PHASE_FILE" 7200 "gardener_phase_callback"
FINAL_PHASE=$(read_phase)
log "Final phase: ${FINAL_PHASE:-none}"
if [ "$FINAL_PHASE" != "PHASE:done" ]; then
case "${_MONITOR_LOOP_EXIT:-}" in
idle_prompt)
log "gardener-agent: Claude returned to prompt without writing phase signal — no phase file written"
;;
idle_timeout)
log "gardener-agent: timed out after 2h with no phase signal"
;;
*)
log "gardener-agent finished without PHASE:done (phase: ${FINAL_PHASE:-none}, exit: ${_MONITOR_LOOP_EXIT:-})"
;;
esac
exit 0
fi
log "claude finished — parsing result file"
# ── Parse result file ─────────────────────────────────────────────────────
CLAUDE_OUTPUT=""
if [ -s "$RESULT_FILE" ]; then
CLAUDE_OUTPUT=$(cat "$RESULT_FILE")
fi
# ── Parse escalations ─────────────────────────────────────────────────────
ESCALATION=$(echo "$CLAUDE_OUTPUT" | awk '/^ESCALATE$/{found=1;next} found && /^(ACTION:|DUST:|CLEAN|PHASE:)/{found=0} found{print}' || true)
if [ -z "$ESCALATION" ]; then
ESCALATION=$(echo "$CLAUDE_OUTPUT" | grep -A50 "^ESCALATE" | grep -E '^[0-9]' || true)
fi
if [ -n "$ESCALATION" ]; then
ITEM_COUNT=$(echo "$ESCALATION" | grep -c '.' || true)
log "Escalating $ITEM_COUNT items to human"
# Send via Matrix (threaded — replies route back via listener)
matrix_send "gardener" "🌱 Issue Gardener — ${ITEM_COUNT} item(s) need attention
${ESCALATION}
Reply with numbers+letters (e.g. 1a 2c) to decide." 2>/dev/null || true
fi
# ── Log actions taken ─────────────────────────────────────────────────────
ACTIONS=$(echo "$CLAUDE_OUTPUT" | grep "^ACTION:" || true)
if [ -n "$ACTIONS" ]; then
echo "$ACTIONS" | while read -r line; do
log " $line"
done
fi
# ── Log dust items (bundling handled by run-gardener formula step) ────
DUST_LINES=$(echo "$CLAUDE_OUTPUT" | grep "^DUST: " || true)
if [ -n "$DUST_LINES" ]; then
DUST_COUNT=$(echo "$DUST_LINES" | grep -c '.' || true)
log "Dust items reported: $DUST_COUNT (bundling handled by run-gardener formula)"
fi
# ── Cleanup scratch file on normal exit ──────────────────────────────────
if [ "$FINAL_PHASE" = "PHASE:done" ]; then
rm -f "$SCRATCH_FILE"
fi
log "--- gardener-agent done ---"

View file

@ -1,138 +0,0 @@
#!/usr/bin/env bash
# =============================================================================
# gardener-poll.sh — Cron wrapper for the gardener agent
#
# Cron: daily (or 2x/day). Handles lock management, escalation reply
# injection, and delegates backlog grooming to gardener-agent.sh.
#
# Grooming (delegated to gardener-agent.sh):
# - Duplicate titles / overlapping scope
# - Missing acceptance criteria
# - Stale issues (no activity > 14 days)
# - Blockers starving the factory
# - Tech-debt promotion / dust bundling
# =============================================================================
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
# Load shared environment (with optional project TOML override)
# Usage: gardener-poll.sh [projects/harb.toml]
export PROJECT_TOML="${1:-}"
# shellcheck source=../lib/env.sh
source "$FACTORY_ROOT/lib/env.sh"
LOG_FILE="$SCRIPT_DIR/gardener.log"
LOCK_FILE="/tmp/gardener-poll.lock"
log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
# ── Lock ──────────────────────────────────────────────────────────────────
if [ -f "$LOCK_FILE" ]; then
LOCK_PID=$(cat "$LOCK_FILE" 2>/dev/null || true)
if [ -n "$LOCK_PID" ] && kill -0 "$LOCK_PID" 2>/dev/null; then
log "poll: gardener running (PID $LOCK_PID)"
exit 0
fi
rm -f "$LOCK_FILE"
fi
echo $$ > "$LOCK_FILE"
trap 'rm -f "$LOCK_FILE"' EXIT
log "--- Gardener poll start ---"
# Gitea labels API requires []int64 — look up the "backlog" label ID once
# Falls back to the known Codeberg repo ID if the API call fails
BACKLOG_LABEL_ID=$(codeberg_api GET "/labels" 2>/dev/null \
| jq -r '.[] | select(.name == "backlog") | .id' 2>/dev/null || true)
BACKLOG_LABEL_ID="${BACKLOG_LABEL_ID:-1300815}"
# ── Check for escalation replies from Matrix ──────────────────────────────
ESCALATION_REPLY=""
if [ -s /tmp/gardener-escalation-reply ]; then
_raw_reply=$(cat /tmp/gardener-escalation-reply)
rm -f /tmp/gardener-escalation-reply
log "Got escalation reply: $(echo "$_raw_reply" | head -1)"
# Filter stale escalation entries referencing already-closed issues (#289).
# Escalation records can persist after the underlying issue resolves; acting
# on them wastes cycles (e.g. creating investigation issues for merged PRs).
while IFS= read -r _reply_line; do
[ -z "$_reply_line" ] && continue
_esc_nums=$(echo "$_reply_line" | grep -oP '#\K\d+' | sort -u || true)
if [ -n "$_esc_nums" ]; then
_any_open=false
for _esc_n in $_esc_nums; do
_esc_st=$(codeberg_api GET "/issues/${_esc_n}" 2>/dev/null \
| jq -r '.state // "open"' 2>/dev/null || echo "open")
if [ "$_esc_st" != "closed" ]; then
_any_open=true
break
fi
done
if [ "$_any_open" = false ]; then
log "Discarding stale escalation (all referenced issues closed): $(echo "$_reply_line" | head -c 120)"
continue
fi
fi
ESCALATION_REPLY="${ESCALATION_REPLY}${_reply_line}
"
done <<< "$_raw_reply"
if [ -n "$ESCALATION_REPLY" ]; then
log "Escalation reply after filtering: $(echo "$ESCALATION_REPLY" | grep -c '.' || echo 0) line(s)"
else
log "All escalation entries were stale — discarded"
fi
fi
export ESCALATION_REPLY
# ── Inject human replies into needs_human dev sessions (backup to supervisor) ─
HUMAN_REPLY_FILE="/tmp/dev-escalation-reply"
for _gr_phase_file in /tmp/dev-session-"${PROJECT_NAME}"-*.phase; do
[ -f "$_gr_phase_file" ] || continue
_gr_phase=$(head -1 "$_gr_phase_file" 2>/dev/null | tr -d '[:space:]' || true)
[ "$_gr_phase" = "PHASE:needs_human" ] || continue
_gr_issue=$(basename "$_gr_phase_file" .phase)
_gr_issue="${_gr_issue#dev-session-${PROJECT_NAME}-}"
[ -z "$_gr_issue" ] && continue
_gr_session="dev-${PROJECT_NAME}-${_gr_issue}"
tmux has-session -t "$_gr_session" 2>/dev/null || continue
# Atomic claim — only take the file once we know a session needs it
_gr_claimed="/tmp/dev-escalation-reply.gardener.$$"
[ -s "$HUMAN_REPLY_FILE" ] && mv "$HUMAN_REPLY_FILE" "$_gr_claimed" 2>/dev/null || continue
_gr_reply=$(cat "$_gr_claimed")
_gr_inject_msg="Human reply received for issue #${_gr_issue}:
${_gr_reply}
Instructions:
1. Read the human's guidance carefully.
2. Continue your work based on their input.
3. When done, push your changes and write the appropriate phase."
_gr_tmpfile=$(mktemp /tmp/human-inject-XXXXXX)
printf '%s' "$_gr_inject_msg" > "$_gr_tmpfile"
tmux load-buffer -b "human-inject-${_gr_issue}" "$_gr_tmpfile" || true
tmux paste-buffer -t "$_gr_session" -b "human-inject-${_gr_issue}" || true
sleep 0.5
tmux send-keys -t "$_gr_session" "" Enter || true
tmux delete-buffer -b "human-inject-${_gr_issue}" 2>/dev/null || true
rm -f "$_gr_tmpfile" "$_gr_claimed"
rm -f "/tmp/dev-renotify-${PROJECT_NAME}-${_gr_issue}"
log "${PROJECT_NAME}: #${_gr_issue} human reply injected into session ${_gr_session} (gardener)"
break # only one reply to deliver
done
# ── Backlog grooming (delegated to gardener-agent.sh) ────────────────────
log "Invoking gardener-agent.sh for backlog grooming"
bash "$SCRIPT_DIR/gardener-agent.sh" "${1:-}" || log "WARNING: gardener-agent.sh exited with error"
log "--- Gardener poll done ---"

View file

@ -1,75 +1,103 @@
#!/usr/bin/env bash
# =============================================================================
# gardener-run.sh — Cron wrapper: files action issue for run-gardener formula
# gardener-run.sh — Cron wrapper: gardener execution via Claude + formula
#
# Runs 2x/day (or on-demand). Guards against concurrent runs and low memory.
# Files an action issue referencing formulas/run-gardener.toml; the action-agent
# picks it up and executes the gardener steps in an interactive Claude session.
# Runs 4x/day (or on-demand). Guards against concurrent runs and low memory.
# Creates a tmux session with Claude (sonnet) reading formulas/run-gardener.toml.
# No action issues — the gardener is a nervous system component, not work (AD-001).
#
# Usage:
# gardener-run.sh [projects/disinto.toml] # project config (default: disinto)
#
# Cron: 0 0,6,12,18 * * * cd /home/debian/dark-factory && bash gardener/gardener-run.sh projects/disinto.toml
# =============================================================================
set -euo pipefail
FACTORY_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
# Load shared environment (with optional project TOML override)
# Usage: gardener-run.sh [projects/harb.toml]
export PROJECT_TOML="${1:-}"
# Accept project config from argument; default to disinto
export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
# shellcheck source=../lib/env.sh
source "$FACTORY_ROOT/lib/env.sh"
# shellcheck source=../lib/file-action-issue.sh
source "$FACTORY_ROOT/lib/file-action-issue.sh"
# 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"
LOG_FILE="$FACTORY_ROOT/gardener/gardener.log"
LOCK_FILE="/tmp/gardener-run.lock"
LOG_FILE="$SCRIPT_DIR/gardener.log"
# shellcheck disable=SC2034 # consumed by run_formula_and_monitor
SESSION_NAME="gardener-${PROJECT_NAME}"
PHASE_FILE="/tmp/gardener-session-${PROJECT_NAME}.phase"
# shellcheck disable=SC2034 # read by monitor_phase_loop in lib/agent-session.sh
PHASE_POLL_INTERVAL=15
SCRATCH_FILE="/tmp/gardener-${PROJECT_NAME}-scratch.md"
RESULT_FILE="/tmp/gardener-result-${PROJECT_NAME}.txt"
log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
# ── Lock ──────────────────────────────────────────────────────────────────
if [ -f "$LOCK_FILE" ]; then
LOCK_PID=$(cat "$LOCK_FILE" 2>/dev/null || true)
if [ -n "$LOCK_PID" ] && kill -0 "$LOCK_PID" 2>/dev/null; then
log "run: gardener-run running (PID $LOCK_PID)"
exit 0
fi
rm -f "$LOCK_FILE"
fi
echo $$ > "$LOCK_FILE"
trap 'rm -f "$LOCK_FILE"' EXIT
# ── Memory guard ──────────────────────────────────────────────────────────
AVAIL_MB=$(free -m | awk '/Mem:/{print $7}')
if [ "${AVAIL_MB:-0}" -lt 2000 ]; then
log "run: skipping — only ${AVAIL_MB}MB available (need 2000)"
exit 0
fi
# ── Guards ────────────────────────────────────────────────────────────────
acquire_cron_lock "/tmp/gardener-run.lock"
check_memory 2000
log "--- Gardener run start ---"
# ── File action issue for run-gardener formula ────────────────────────────
ISSUE_BODY="---
formula: run-gardener
model: opus
---
# ── Consume escalation replies ────────────────────────────────────────────
consume_escalation_reply "gardener"
Periodic gardener housekeeping run. The action-agent reads \`formulas/run-gardener.toml\`
and executes the steps: preflight, grooming, blocked-review,
AGENTS.md update, and commit-and-pr.
# ── Load formula + context ───────────────────────────────────────────────
load_formula "$FACTORY_ROOT/formulas/run-gardener.toml"
build_context_block AGENTS.md
Filed automatically by \`gardener-run.sh\`."
# ── Read scratch file (compaction survival) ───────────────────────────────
SCRATCH_CONTEXT=$(read_scratch_context "$SCRATCH_FILE")
SCRATCH_INSTRUCTION=$(build_scratch_instruction "$SCRATCH_FILE")
_rc=0
file_action_issue "run-gardener" "action: run-gardener — periodic housekeeping" "$ISSUE_BODY" || _rc=$?
case "$_rc" in
0) ;;
1) log "run: open run-gardener action issue already exists — skipping"
log "--- Gardener run done ---"
exit 0 ;;
2) log "ERROR: 'action' label not found — cannot file gardener issue"
exit 1 ;;
*) log "ERROR: failed to create action issue for run-gardener"
exit 1 ;;
esac
# ── Build prompt (gardener needs extra API endpoints for issue management) ─
GARDENER_API_EXTRA="
Relabel: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X PUT -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}/labels' -d '{\"labels\":[LABEL_ID]}'
Comment: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X POST -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}/comments' -d '{\"body\":\"...\"}'
Close: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X PATCH -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}' -d '{\"state\":\"closed\"}'
Edit body: curl -sf -H \"Authorization: token \$CODEBERG_TOKEN\" -X PATCH -H 'Content-Type: application/json' '${CODEBERG_API}/issues/{number}' -d '{\"body\":\"new body\"}'
"
build_prompt_footer "$GARDENER_API_EXTRA"
log "Filed action issue #${FILED_ISSUE_NUM} for run-gardener formula"
matrix_send "gardener" "Filed action #${FILED_ISSUE_NUM}: run-gardener — periodic housekeeping" 2>/dev/null || true
# shellcheck disable=SC2034 # consumed by run_formula_and_monitor
PROMPT="You are the issue gardener for ${CODEBERG_REPO}. Work through the formula below. You MUST write PHASE:done to '${PHASE_FILE}' when finished — the orchestrator will time you out if you return to the prompt without signalling.
log "--- Gardener run done ---"
You have full shell access and --dangerously-skip-permissions.
Fix what you can. Escalate what you cannot. Do NOT ask permission — act first, report after.
${ESCALATION_REPLY:+
## Escalation Reply (from Matrix — human message)
${ESCALATION_REPLY}
Act on this reply during the grooming step.
}
## Project context
${CONTEXT_BLOCK}
${SCRATCH_CONTEXT:+${SCRATCH_CONTEXT}
}
## Result file
Write actions and dust items to: ${RESULT_FILE}
## Formula
${FORMULA_CONTENT}
${SCRATCH_INSTRUCTION}
${PROMPT_FOOTER}"
# ── Reset result file ────────────────────────────────────────────────────
rm -f "$RESULT_FILE"
touch "$RESULT_FILE"
# ── Run session ──────────────────────────────────────────────────────────
export CLAUDE_MODEL="sonnet"
run_formula_and_monitor "gardener" 7200
# ── Cleanup scratch file on normal exit ──────────────────────────────────
# FINAL_PHASE already set by run_formula_and_monitor
if [ "${FINAL_PHASE:-}" = "PHASE:done" ]; then
rm -f "$SCRATCH_FILE"
fi

View file

@ -12,7 +12,7 @@ sourced as needed.
| `lib/load-project.sh` | Parses a `projects/*.toml` file into env vars (`PROJECT_NAME`, `CODEBERG_REPO`, `WOODPECKER_REPO_ID`, monitoring toggles, Matrix config, etc.). | env.sh (when `PROJECT_TOML` is set), supervisor-poll (per-project iteration) |
| `lib/parse-deps.sh` | Extracts dependency issue numbers from an issue body (stdin → stdout, one number per line). Matches `## Dependencies` / `## Depends on` / `## Blocked by` sections and inline `depends on #N` patterns. Not sourced — executed via `bash lib/parse-deps.sh`. | dev-poll, supervisor-poll |
| `lib/matrix_listener.sh` | Long-poll Matrix sync daemon. Dispatches thread replies to the correct agent via well-known files (`/tmp/{agent}-escalation-reply`). Handles supervisor, gardener, dev, review, vault, and action reply routing. Run as systemd service. | Standalone daemon |
| `lib/formula-session.sh` | `acquire_cron_lock()`, `check_memory()`, `load_formula()`, `build_context_block()`, `start_formula_session()`, `formula_phase_callback()`, `build_prompt_footer()`, `run_formula_and_monitor()` — shared helpers for formula-driven cron agents (lock, memory guard, formula loading, prompt assembly, tmux session, monitor loop, crash recovery). | planner-run.sh, predictor-run.sh |
| `lib/formula-session.sh` | `acquire_cron_lock()`, `check_memory()`, `load_formula()`, `build_context_block()`, `consume_escalation_reply()`, `start_formula_session()`, `formula_phase_callback()`, `build_prompt_footer()`, `run_formula_and_monitor()` — shared helpers for formula-driven cron agents (lock, memory guard, formula loading, prompt assembly, tmux session, monitor loop, crash recovery). | planner-run.sh, predictor-run.sh, gardener-run.sh, supervisor-run.sh |
| `lib/secret-scan.sh` | `scan_for_secrets()` — detects potential secrets (API keys, bearer tokens, private keys, URLs with embedded credentials) in text; returns 1 if secrets found. `redact_secrets()` — replaces detected secret patterns with `[REDACTED]`. | file-action-issue.sh, phase-handler.sh |
| `lib/file-action-issue.sh` | `file_action_issue()` — dedup check, secret scan, label lookup, and issue creation for formula-driven cron wrappers. Sets `FILED_ISSUE_NUM` on success. Returns 4 if secrets detected in body. | gardener-run.sh |
| `lib/agent-session.sh` | Shared tmux + Claude session helpers: `create_agent_session()`, `inject_formula()`, `agent_wait_for_claude_ready()`, `agent_inject_into_session()`, `agent_kill_session()`, `monitor_phase_loop()`, `read_phase()`, `write_compact_context()`. `create_agent_session(session, workdir, [phase_file])` optionally installs a PostToolUse hook (matcher `Bash\|Write`) that detects phase file writes in real-time — when Claude writes to the phase file, the hook writes a marker so `monitor_phase_loop` reacts on the next poll instead of waiting for mtime changes. Also installs a StopFailure hook (matcher `rate_limit\|server_error\|authentication_failed\|billing_error`) that writes `PHASE:failed` with an `api_error` reason to the phase file and touches the phase-changed marker, so the orchestrator discovers API errors within one poll cycle instead of waiting for idle timeout. Also installs a SessionStart hook (matcher `compact`) that re-injects phase protocol instructions after context compaction — callers write the context file via `write_compact_context(phase_file, content)`, and the hook (`on-compact-reinject.sh`) outputs the file content to stdout so Claude retains critical instructions. When `MATRIX_THREAD_ID` is exported, also installs a Stop hook (`on-stop-matrix.sh`) that streams each Claude response to the Matrix thread. `monitor_phase_loop` sets `_MONITOR_LOOP_EXIT` to one of: `done`, `idle_timeout`, `idle_prompt` (Claude returned to `>` for 3 consecutive polls without writing any phase — callback invoked with `PHASE:failed`, session already dead), `crashed`, or a `PHASE:*` string. **Callers must handle `idle_prompt`** in both their callback and their post-loop exit handler — see [`docs/PHASE-PROTOCOL.md` idle_prompt](docs/PHASE-PROTOCOL.md#idle_prompt-exit-reason) for the full contract. | dev-agent.sh, gardener-agent.sh, action-agent.sh |
| `lib/file-action-issue.sh` | `file_action_issue()` — dedup check, secret scan, label lookup, and issue creation for formula-driven cron wrappers. Sets `FILED_ISSUE_NUM` on success. Returns 4 if secrets detected in body. | (available for future use) |
| `lib/agent-session.sh` | Shared tmux + Claude session helpers: `create_agent_session()`, `inject_formula()`, `agent_wait_for_claude_ready()`, `agent_inject_into_session()`, `agent_kill_session()`, `monitor_phase_loop()`, `read_phase()`, `write_compact_context()`. `create_agent_session(session, workdir, [phase_file])` optionally installs a PostToolUse hook (matcher `Bash\|Write`) that detects phase file writes in real-time — when Claude writes to the phase file, the hook writes a marker so `monitor_phase_loop` reacts on the next poll instead of waiting for mtime changes. Also installs a StopFailure hook (matcher `rate_limit\|server_error\|authentication_failed\|billing_error`) that writes `PHASE:failed` with an `api_error` reason to the phase file and touches the phase-changed marker, so the orchestrator discovers API errors within one poll cycle instead of waiting for idle timeout. Also installs a SessionStart hook (matcher `compact`) that re-injects phase protocol instructions after context compaction — callers write the context file via `write_compact_context(phase_file, content)`, and the hook (`on-compact-reinject.sh`) outputs the file content to stdout so Claude retains critical instructions. When `MATRIX_THREAD_ID` is exported, also installs a Stop hook (`on-stop-matrix.sh`) that streams each Claude response to the Matrix thread. `monitor_phase_loop` sets `_MONITOR_LOOP_EXIT` to one of: `done`, `idle_timeout`, `idle_prompt` (Claude returned to `>` for 3 consecutive polls without writing any phase — callback invoked with `PHASE:failed`, session already dead), `crashed`, or a `PHASE:*` string. **Callers must handle `idle_prompt`** in both their callback and their post-loop exit handler — see [`docs/PHASE-PROTOCOL.md` idle_prompt](docs/PHASE-PROTOCOL.md#idle_prompt-exit-reason) for the full contract. | dev-agent.sh, action-agent.sh |

View file

@ -2,7 +2,7 @@
# formula-session.sh — Shared helpers for formula-driven cron agents
#
# Provides reusable functions for the common cron-wrapper + tmux-session
# pattern used by planner-run.sh and gardener-agent.sh.
# pattern used by planner-run.sh, predictor-run.sh, gardener-run.sh, and supervisor-run.sh.
#
# Functions:
# acquire_cron_lock LOCK_FILE — PID lock with stale cleanup
@ -81,6 +81,25 @@ $(cat "$ctx_path")
done
}
# ── Escalation reply consumption ─────────────────────────────────────────
# consume_escalation_reply AGENT_NAME
# Atomically consumes /tmp/{agent}-escalation-reply if it exists.
# Sets ESCALATION_REPLY to the file contents (empty string if no reply).
consume_escalation_reply() {
local agent="$1"
local reply_file="/tmp/${agent}-escalation-reply"
ESCALATION_REPLY=""
if [ -s "$reply_file" ]; then
local tmp_file="${reply_file}.consumed.$$"
if mv "$reply_file" "$tmp_file" 2>/dev/null; then
ESCALATION_REPLY=$(cat "$tmp_file")
rm -f "$tmp_file"
log "Consumed escalation reply: $(echo "$ESCALATION_REPLY" | head -1)"
fi
fi
}
# ── Session management ───────────────────────────────────────────────────
# start_formula_session SESSION WORKDIR PHASE_FILE

View file

@ -57,16 +57,7 @@ else
fi
# ── Consume escalation replies ────────────────────────────────────────────
# Move the file atomically so matrix_listener can write a new one
ESCALATION_REPLY=""
if [ -s /tmp/supervisor-escalation-reply ]; then
_reply_tmp="/tmp/supervisor-escalation-reply.consumed.$$"
if mv /tmp/supervisor-escalation-reply "$_reply_tmp" 2>/dev/null; then
ESCALATION_REPLY=$(cat "$_reply_tmp")
rm -f "$_reply_tmp"
log "Consumed escalation reply: $(echo "$ESCALATION_REPLY" | head -1)"
fi
fi
consume_escalation_reply "supervisor"
# ── Load formula + context ───────────────────────────────────────────────
load_formula "$FACTORY_ROOT/formulas/run-supervisor.toml"