fix(cloud): gh auth setup-git + --repo flag + push failure tracking

3 critical bugs fixed in agent entrypoint:
1. gh auth setup-git — bridges gh→git credential helper (push was failing)
2. --repo flag on all gh issue/pr commands — bare-repo worktrees lack remote context
3. PUSH_OK tracking — push failures now reported instead of silently swallowed

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Antigravity Agent

.trinity/mu/heartbeat.json

@@ -1 +1 @@
-{"agent":"mu","wake":74,"timestamp":1773238120,"errors_scanned":25,"fixes_applied":0,"build_ok":false,"test_ok":true}
+{"agent":"mu","wake":75,"timestamp":1773248114,"errors_scanned":25,"fixes_applied":0,"build_ok":false,"test_ok":true}

.trinity/mu/state/wake_count

@@ -1 +1 @@
-74
+75

.trinity/night-evolution-map.md

@@ -1,79 +1,96 @@
 # Night Evolution Map — Cloud Dev Pipeline Hardening
 # 2026-03-11 night → 2026-03-12 morning
 
-## Current State (updated)
-- 3 PRs merged (#129, #130, #138), 3 closed with conflicts (#132, #133, #139)
-- PR #141 (Golden Chain pipeline from agent-140) — CI pending, compile fixes pushed
-- Docker image rebuilt with heartbeat + pipefail + Telegram fixes
-- 2 Railway services reusable (agent-126, agent-131)
-- agent-126 redeployed for issue #134, agent-131 for issue #135
+## Final Status
 
-## Evolution Phases (Priority Order)
+### PRs Merged (5) + Closed (2 quality)
+| PR | Title | Source |
+|----|-------|--------|
+| #129 | JSONL event persistence + deduplication | agent-124 |
+| #130 | Git worktree isolation for faster startup | agent-125 |
+| #138 | Buffer size increase, CLAUDE.md.agent to .gitignore | agent-136 |
+| #141 | Golden Chain pipeline — tri cloud pipeline/verify/merge | agent-140 |
+| #142 | Telegram log streaming — batch every 5s + output classifier | agent-131 |
 
-### Phase 1: Merge Ready PRs ✅ DONE
-- [x] PR #129 (JSONL event persistence) → merged
-- [x] PR #130 (git worktree isolation) → merged
-- [x] PR #138 (fix #136 — buffer + .gitignore) → merged
-- [x] PR #132, #133, #139 → closed (conflicts after merges)
-- [x] PR #141 (agent-140 Golden Chain) → compile fixes pushed, CI pending
+### PRs Closed (3, superseded by direct fixes)
+| PR | Reason |
+|----|--------|
+| #132 | Merge conflict after #129/#130 merged |
+| #133 | Merge conflict after #129/#130 merged |
+| #139 | Merge conflict, fixes applied directly |
 
-### Phase 2: Entrypoint Hardening ✅ DONE
-- [x] Heartbeat subshell bug → temp file `/tmp/agent_heartbeat_state`
-- [x] `report_status()` writes to heartbeat file
-- [x] Telegram notification ordering (BEFORE LAST_STATUS update)
-- [x] `set -eo pipefail` + `#!/bin/bash` shebang
-- [x] HTML escape helper `escape_html()`
-- [x] `send_telegram()` uses temp file for JSON (no escaping issues)
-- [x] Docker image rebuilt + pushed to GHCR
+### Issues Closed (5)
+| Issue | Resolution |
+|-------|-----------|
+| #134 | Fixed: u32→i64 timestamp, entry_idx dedup |
+| #135 | Fixed: VOLUME shadow, worktree -b branch |
+| #136 | Fixed via PR #138 merge |
+| #137 | Fixed: pipefail, bash shebang, Telegram ordering |
+| #140 | Fixed via PR #141 merge |
 
-### Phase 3: Orchestrator CLI (partially done by agent-140)
-Agent-140's PR #141 adds:
-- [x] `tri cloud pipeline <N>` — spawn → monitor → verify → merge → cleanup
-- [x] `tri cloud verify <N>` — local zig build check
-- [x] `tri cloud merge <N>` — merge PR via gh CLI
-- [x] Enhanced `tri cloud agents` — stuck detection, health indicators, elapsed formatting
-Already working from before:
-- [x] `tri cloud spawn <N>` — calls Railway API
-- [x] `tri cloud kill <N>` — delete service
-- [x] `tri cloud agents` — list active containers
-- [ ] `tri cloud logs <N>` — fetch Railway deploy logs
-- [ ] Service recycling in CLI (currently manual via env var update)
+### Direct Commits to Main (3)
+1. `b470c5ae7` — heartbeat subshell + pipefail + Telegram ordering + HTML escape
+2. `fe6dc534e` — u32 overflow, entry_idx duplicates, VOLUME shadow, worktree conflict
+3. Merge commits for PRs #129, #130, #138, #141
 
-### Phase 4: Auto-Pipeline (in PR #141)
-- [x] Spawn → monitor heartbeats → detect DONE/FAIL (in PR)
-- [x] On DONE: fetch PR, run `zig build` locally (in PR)
-- [x] On pass: auto-merge PR (in PR)
-- [x] On fail: respawn (max 3x) (in PR)
-- [ ] Create fix-issue with review on failure
-- [ ] Cleanup container after completion
+### Docker Image Rebuilt (2x)
+- First: heartbeat + pipefail + Telegram fixes
+- Second: VOLUME shadow removal + worktree branch fix
 
-### Phase 5: Monitoring & Metrics
-- [x] JSONL event persistence (PR #129 merged)
-- [ ] Agent solve rate dashboard
-- [ ] Cost per agent tracking
-- [ ] Token usage estimation
-- [ ] Success/fail/retry counters
+## Phase Completion
 
-### Phase 6: Agent Intelligence
-- [x] Agent reads CLAUDE.md (via SOUL.md injection)
-- [x] Better commit messages (include issue number)
-- [ ] Agent checks out existing branch for fix-issues
-- [ ] Agent runs `zig build -Dci=true` instead of full build
-- [ ] Multi-file context awareness
+| Phase | Status | Detail |
+|-------|--------|--------|
+| 1. Merge PRs | DONE | 4 merged, 3 closed |
+| 2. Entrypoint Hardening | DONE | 6 fixes applied |
+| 3. Orchestrator CLI | 80% | pipeline/verify/merge added, logs TBD |
+| 4. Auto-Pipeline | 70% | In PR #141, needs testing |
+| 5. Monitoring | 30% | JSONL working, dashboard TBD |
+| 6. Agent Intelligence | 20% | SOUL.md works, branch reuse TBD |
 
-## Active Agents
-- agent-126 → issue #134 (fix PR #129 bugs — u32 timestamp, duplicates, buffer)
-- agent-131 → issue #135 (fix PR #130 bugs — VOLUME shadow, worktree conflicts)
+## Remaining Open Issues
+- #131 feat(cloud): Stream all container logs to Telegram in realtime
+- #126 Cloud Dev: Structured ACI protocol
+- #128, #127 FPGA/pipeline TODOs (lower priority)
 
-## Next Steps
-1. Wait for PR #141 CI → merge if passes
-2. Monitor agents #134, #135 → review PRs when ready
-3. Spawn agent for #137 (fix PR #133 bugs) when slot frees up
-4. Create issue for `tri cloud logs` command
-5. Create issue for service recycling in CLI
+## Key Fixes Applied
+1. Heartbeat reads from temp file (subshell isolation solved)
+2. Telegram gets notifications on every status change (ordering fix)
+3. HTML escaping + safe JSON via temp files
+4. `#!/bin/bash` + `set -eo pipefail`
+5. `i64` timestamps (no more u32 overflow)
+6. No duplicate JSONL entries
+7. No VOLUME shadowing bare repo
+8. Concurrent agents get unique branches
+9. Golden Chain: `tri cloud pipeline <N>` automates full cycle
+10. Telegram `editMessageText` — 1 dashboard message updated in place
+11. `NO_COLOR=1` in containers for clean output
+12. Worktree lock/unlock prevents accidental pruning
+13. Workflow reuses services instead of delete+create (avoids 25/day limit)
 
-## Constraints
-- 2 Railway services available (agent-126, agent-131)
-- z.ai proxy ~8min per agent run
-- Telegram 30 msg/min rate limit
-- Docker rebuild ~90s (cached layers)
+## Active Agents (latest cycle — 16:33 UTC)
+- **ubuntu** service → #126 — 🔴 FAILED (0 commits, 619s — issue too abstract for autonomous agent)
+- **Agents Anywhere** service → #131 — 🔵 DONE → PR #142 merged
+- **Agents Anywhere** service → #115 (VIBEE eqlPrimitive fix) — 🔴 DONE but push failed 3x, no PR created
+- **ubuntu** service → #114 (VIBEE undefined Field type) — 🔴 DONE but push failed (git auth bug)
+- **Agents Anywhere** service → #116 (Re-verify stale ast-check) — 🔴 FAILED (gh can't read issue — missing --repo)
+- PR #143 from agent-126 — 🔴 CLOSED (review: grep -oP not portable, worktree cleanup order)
+- **Docker rebuild #3** — fixes: `gh auth setup-git`, `--repo` on all gh commands, PUSH_OK tracking
+- **ubuntu** service → #114 (RETRY) — 🚀 REDEPLOYED 16:55 UTC with fixed image
+- **Agents Anywhere** service → #116 (RETRY) — 🚀 REDEPLOYED 16:55 UTC with fixed image
+
+## Bug Found & Fixed This Cycle
+14. `sleepApplication: true` on "Agents Anywhere" service — Railway was sleeping container before entrypoint ran. Fixed via `serviceInstanceUpdate` + redeploy.
+
+## Lessons Learned
+1. Railway MCP `deploy` uploads source, NOT Docker image — use GraphQL API
+2. `startCommand` overrides Docker ENTRYPOINT — must set via serviceInstanceUpdate
+3. 25 service/day creation limit — never delete+create, always reuse
+4. `variableCollectionUpsert` needs actual values, not empty shell vars
+5. Service names with spaces break Railway CLI — avoid spaces in service names
+6. `sleepApplication: true` silently kills agent containers — always set to false for batch jobs
+7. Abstract/design issues (#126 "Structured ACI protocol") produce 0 commits — agents need concrete, code-level tasks with specific files/functions to modify
+8. `retry "git push ... 2>/dev/null" || true` silently swallows push failures — agent reports DONE with no PR. Fixed: track PUSH_OK, skip PR creation if push fails, report FAILED explicitly
+9. **CRITICAL**: `gh auth login` only configures `gh` CLI, NOT `git push`. Fixed: `gh auth setup-git`
+10. **CRITICAL**: All `gh issue/pr` commands lack `--repo` flag — bare-repo worktrees have no git remote context. Fixed: extract `GH_REPO` from `REPO_URL`, add `--repo` to all gh calls
+11. Docker rebuild #3 deployed with fixes #8-10. Both services redeployed 16:55 UTC

deploy/agent-entrypoint.sh

@@ -8,6 +8,8 @@
 set -eo pipefail
 
 REPO_URL="${REPO_URL:-https://github.com/gHashTag/trinity.git}"
+# Extract owner/repo for gh --repo flag (bare-repo worktrees lack git remote context)
+GH_REPO=$(echo "${REPO_URL}" | sed 's|.*github.com[:/]||; s|\.git$||')
 ISSUE="${ISSUE_NUMBER:?ISSUE_NUMBER is required}"
 AGENT_TIMEOUT="${AGENT_TIMEOUT:-3600}"  # 1 hour default
 HEARTBEAT_INTERVAL="${HEARTBEAT_INTERVAL:-30}"
@@ -76,7 +78,7 @@ ${CURRENT_DETAIL} (${ELAPSED}s)"
 
     # 3. GitHub issue comment on status change (skip duplicates)
     if [ "${CURRENT_STATUS}" != "${LAST_STATUS}" ]; then
-        gh issue comment "${ISSUE}" --body "${EMOJI} **Trinity Agent** | ${TIMESTAMP}
+        gh issue comment "${ISSUE}" --repo "${GH_REPO}" --body "${EMOJI} **Trinity Agent** | ${TIMESTAMP}
 📋 **Step**: ${STEP_NUM}/${TOTAL_STEPS} — ${CURRENT_DETAIL}
 🔄 **Status**: ${CURRENT_STATUS}
 ⏱️ **Elapsed**: ${ELAPSED}s" 2>/dev/null || log "Warning: GitHub comment failed"
@@ -96,7 +98,7 @@ ${CURRENT_DETAIL} (${ELAPSED}s)"
 | **Updated** | ${TIMESTAMP} |"
 
     if [ -z "${DASHBOARD_COMMENT_ID}" ]; then
-        DASHBOARD_COMMENT_ID=$(gh issue comment "${ISSUE}" --body "${DASHBOARD_BODY}" 2>/dev/null | grep -o '/[0-9]*$' | tr -d '/' || true)
+        DASHBOARD_COMMENT_ID=$(gh issue comment "${ISSUE}" --repo "${GH_REPO}" --body "${DASHBOARD_BODY}" 2>/dev/null | grep -o '/[0-9]*$' | tr -d '/' || true)
         if [ -z "${DASHBOARD_COMMENT_ID}" ]; then
             DASHBOARD_COMMENT_ID=$(gh api "repos/{owner}/{repo}/issues/${ISSUE}/comments" --jq '.[-1].id' 2>/dev/null || true)
         fi
@@ -378,6 +380,10 @@ log "gh auth status: ${GH_STATUS}"
 git config --global user.name "Trinity Agent"
 git config --global user.email "trinity-agent@users.noreply.github.com"
 
+# Configure git to use gh as credential helper (fixes push auth)
+gh auth setup-git 2>/dev/null || true
+log "gh auth setup-git done — git push will use GITHUB_TOKEN"
+
 # === 2. Setup worktree from shared bare repo ===
 report_status "AWAKENING" "Creating worktree from bare repository"
 
@@ -422,7 +428,7 @@ sed "s/{ISSUE_NUMBER}/${ISSUE}/g" /etc/trinity/SOUL.md > "${WORKTREE_PATH}/CLAUD
 
 # === 4. Read issue ===
 report_status "READING" "Reading issue #${ISSUE}"
-ISSUE_BODY=$(gh issue view "${ISSUE}" --json title,body,labels --jq '.' 2>/dev/null || echo '{"title":"Unknown","body":"Failed to fetch issue"}')
+ISSUE_BODY=$(gh issue view "${ISSUE}" --repo "${GH_REPO}" --json title,body,labels --jq '.' 2>/dev/null || echo '{"title":"Unknown","body":"Failed to fetch issue"}')
 ISSUE_TITLE=$(echo "${ISSUE_BODY}" | grep -oP '"title"\s*:\s*"[^"]*"' | head -1 | sed 's/"title"\s*:\s*"//;s/"$//' || echo "issue #${ISSUE}")
 log "Issue title: ${ISSUE_TITLE}"
 send_telegram "📖 <b>Agent #${ISSUE}</b> читает задачу:
@@ -468,7 +474,7 @@ emit_event "command" "{\"cmd\":\"claude\",\"exit_code\":${CLAUDE_EXIT},\"timeout
 
 if [ "${CLAUDE_EXIT}" -eq 124 ]; then
     report_status "STUCK" "Timeout after ${AGENT_TIMEOUT}s"
-    gh issue comment "${ISSUE}" --body "⏰ **Trinity Agent**: Timed out after ${AGENT_TIMEOUT}s. Manual intervention needed." 2>/dev/null || true
+    gh issue comment "${ISSUE}" --repo "${GH_REPO}" --body "⏰ **Trinity Agent**: Timed out after ${AGENT_TIMEOUT}s. Manual intervention needed." 2>/dev/null || true
 elif [ "${CLAUDE_EXIT}" -ne 0 ]; then
     report_status "ERROR" "Claude Code exited with code ${CLAUDE_EXIT}"
 fi
@@ -521,7 +527,7 @@ fi
 # === 8. Push and create PR if not already done ===
 report_status "TESTING" "Checking/creating PR"
 stream_to_telegram "Checking for existing PR..."
-EXISTING_PR=$(gh pr list --head "feat/issue-${ISSUE}" --json number --jq '.[0].number' 2>/dev/null || echo "")
+EXISTING_PR=$(gh pr list --repo "${GH_REPO}" --head "feat/issue-${ISSUE}" --json number --jq '.[0].number' 2>/dev/null || echo "")
 
 if [ -z "${EXISTING_PR}" ]; then
     # Check if there are actually commits to push
@@ -530,12 +536,22 @@ if [ -z "${EXISTING_PR}" ]; then
     if [ "${COMMIT_COUNT}" -gt 0 ]; then
         log "Pushing ${COMMIT_COUNT} commit(s)..."
         stream_to_telegram "Pushing ${COMMIT_COUNT} commit(s) to origin..."
-        retry "git push -u origin 'feat/issue-${ISSUE}' 2>/dev/null" || true
+        PUSH_OK=0
+        retry "git push -u origin 'feat/issue-${ISSUE}'" && PUSH_OK=1 || true
         stream_to_telegram "Push completed."
 
+        if [ "${PUSH_OK}" -eq 0 ]; then
+            log "Push failed after 3 retries — cannot create PR"
+            report_status "FAILED" "Push failed after 3 retries"
+            send_telegram "❌ Agent #${ISSUE}: Push failed after 3 retries — code ready but cannot push"
+            # Still try to report what happened
+            gh issue comment "${ISSUE}" --repo "${GH_REPO}" --body "❌ **Trinity Agent**: Code committed locally but push to origin failed 3 times. Branch: feat/issue-${ISSUE}" 2>/dev/null || true
+        fi
+
+        if [ "${PUSH_OK}" -eq 1 ]; then
         log "Creating PR..."
         stream_to_telegram "Creating pull request..."
-        PR_URL=$(gh pr create \
+        PR_URL=$(gh pr create --repo "${GH_REPO}" \
             --title "feat: solve issue #${ISSUE}" \
             --body "Closes #${ISSUE}
 
@@ -553,7 +569,7 @@ Commits: ${COMMIT_COUNT}" \
             DIFF_STAT=$(git diff --stat main..HEAD 2>/dev/null || echo "N/A")
             FINAL_ELAPSED=$(( $(date +%s) - START_TIME ))
             stream_to_telegram "Posting final summary..."
-            gh issue comment "${ISSUE}" --body "🚀 **Trinity Agent — Summary**
+            gh issue comment "${ISSUE}" --repo "${GH_REPO}" --body "🚀 **Trinity Agent — Summary**
 
 | Field | Value |
 |-------|-------|
@@ -577,10 +593,11 @@ ${DIFF_STAT}
         else
             stream_to_telegram "Failed to create PR."
         fi
+        fi
     else
         stream_to_telegram "No commits produced — agent could not solve issue."
         report_status "FAILED" "No commits produced — agent could not solve issue"
-        gh issue comment "${ISSUE}" --body "❌ **Trinity Agent**: No solution produced. Issue may need manual attention." 2>/dev/null || true
+        gh issue comment "${ISSUE}" --repo "${GH_REPO}" --body "❌ **Trinity Agent**: No solution produced. Issue may need manual attention." 2>/dev/null || true
     fi
 else
     stream_to_telegram "PR already exists: #${EXISTING_PR}"

fpga/openxc7-synth/Makefile.200t

@@ -0,0 +1,69 @@
+# =============================================================================
+# HSLM on Artix-7 XC7A200T — Scaled-up Ternary Transformer
+# =============================================================================
+# 365 BRAM36 (2.7x more than XC7A100T)
+# Allows: 11 TrinityBlocks + Embedding + LM Head + Attention
+#
+# Board options:
+#   - Digilent Nexys A7 200T (~$300)
+#   - AliExpress XC7A200T boards (~$50-80)
+#   - QMTECH 200T variant (if available)
+#
+# Just change PART to synthesize for 200T
+# =============================================================================
+
+FAMILY  = artix7
+PART    = xc7a200tfbg484-1
+PROJECT = trinity_200t
+TOP_MODULE = hslm_full_top
+XDC     = trinity_200t.xdc
+
+# Paths for openXC7 tools (snap-based)
+NEXTPNR_XILINX_DIR ?= /snap/openxc7/current/opt/nextpnr-xilinx
+NEXTPNR_XILINX_PYTHON_DIR ?= $(NEXTPNR_XILINX_DIR)/python
+PRJXRAY_DB_DIR ?= $(NEXTPNR_XILINX_DIR)/external/prjxray-db
+CHIPDB ?= ./chipdb/
+
+DBPART = $(shell echo $(PART) | sed -e 's/-[0-9]//g')
+
+PYPY3 ?= pypy3
+
+# Source files for 11-block design
+VERILOG_SRCS = hslm_full_top.v trinity_block.v ternary_matvec_bram.v \
+               ternary_activation.v ternary_rmsnorm.v embedding_lookup.v \
+               lm_head_matvec.v argmax_unit.v
+
+.PHONY: all clean info
+
+all: $(PROJECT).bit
+
+info:
+	@echo "=== Trinity 200T Build ==="
+	@echo "PART:       $(PART)"
+	@echo "BRAM36:     365 (vs 135 on 100T)"
+	@echo "LUT:        134,600 (vs 63,400)"
+	@echo "DSP48:      740 (vs 240)"
+	@echo "Potential:  11 TrinityBlocks (~1.9M weights)"
+	@echo ""
+
+$(PROJECT).json: $(VERILOG_SRCS)
+	yosys -p "synth_xilinx -flatten -abc9 -arch xc7 -top $(TOP_MODULE); write_json $(PROJECT).json" $(VERILOG_SRCS)
+
+$(CHIPDB)/$(DBPART).bin:
+	mkdir -p $(CHIPDB)
+	$(PYPY3) $(NEXTPNR_XILINX_PYTHON_DIR)/bbaexport.py --device $(PART) --bba $(DBPART).bba
+	bbasm -l $(DBPART).bba $(CHIPDB)/$(DBPART).bin
+	rm -f $(DBPART).bba
+
+$(PROJECT).fasm: $(PROJECT).json $(CHIPDB)/$(DBPART).bin $(XDC)
+	nextpnr-xilinx --chipdb $(CHIPDB)/$(DBPART).bin --xdc $(XDC) --json $(PROJECT).json --fasm $@
+
+$(PROJECT).frames: $(PROJECT).fasm
+	fasm2frames --part $(PART) --db-root $(PRJXRAY_DB_DIR)/$(FAMILY) $< > $@
+
+$(PROJECT).bit: $(PROJECT).frames
+	xc7frames2bit --part_file $(PRJXRAY_DB_DIR)/$(FAMILY)/$(PART)/part.yaml --part_name $(PART) --frm_file $< --output_file $@
+
+clean:
+	rm -f *.bit *.frames *.fasm *.json
+	rm -rf chipdb/