Skip to content

feat(agent): Ralph sleep-wake daemon — native Zig binary#50

Merged
gHashTag merged 1 commit into
mainfrom
ralph/w1/ralph-agent
Mar 9, 2026
Merged

feat(agent): Ralph sleep-wake daemon — native Zig binary#50
gHashTag merged 1 commit into
mainfrom
ralph/w1/ralph-agent

Conversation

@gHashTag
Copy link
Copy Markdown
Owner

@gHashTag gHashTag commented Mar 9, 2026

Summary

  • 8 Zig modules (~500 LOC) implementing a sleep-wake autonomous agent daemon
  • Polls GitHub Issues (assign:ralph,status:pending), builds context from IDENTITY + HANDOVER + state, spawns Claude CLI, manages handover between sessions
  • Native Zig binary: zig build agent / zig build agent -- --single-shot
  • Zero bash — pure Zig using std.http.Client, std.process.Child.run, std.debug.print

Architecture

tools/mcp/trinity_mcp/agent/
├── main.zig              # Entry point + env vars + CLI args
├── agent_loop.zig        # WAKE→WORK→SLEEP cycle
├── identity.zig          # Load .ralph/IDENTITY.md
├── context_builder.zig   # Assemble structured prompt
├── github_poller.zig     # Fetch pending issues via GitHub REST API
├── claude_runner.zig     # Spawn claude CLI as child process
├── handover.zig          # Read/write .ralph/HANDOVER.md
└── state.zig             # File-based key-value persistence

Test plan

  • zig build compiles without errors
  • zig fmt --check passes
  • Binary runs and exits with expected error when GH_TOKEN missing
  • CI Build & Test passes

🤖 Generated with Claude Code

@claude
feat(agent): Ralph sleep-wake daemon — native Zig, GitHub polling, Cl…
d754a2a 
…aude CLI spawn

8 Zig modules (~500 LOC):
- main.zig: entry point, env vars, CLI args
- agent_loop.zig: WAKE→WORK→SLEEP cycle
- identity.zig: load .ralph/IDENTITY.md
- context_builder.zig: assemble structured prompt
- github_poller.zig: fetch pending issues via HTTP
- claude_runner.zig: spawn `claude` CLI as child process
- handover.zig: read/write .ralph/HANDOVER.md
- state.zig: file-based key-value persistence

Build: `zig build agent` / `zig build agent -- --single-shot`

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@gHashTag gHashTag merged commit 1f742cf into main Mar 9, 2026
7 of 9 checks passed
@gHashTag gHashTag deleted the ralph/w1/ralph-agent branch March 9, 2026 12:07
@gHashTag gHashTag mentioned this pull request Mar 15, 2026
Closed
gHashTag pushed a commit that referenced this pull request Mar 15, 2026
@claude
fix(fpga): UG470-compliant verify with frame alignment and ECC/BRAM m…
dfcef5d 
…asking

cmd_verify rewritten to properly compare readback vs .bit file:
- Find FDRI write command in .bit → extract frame data at correct offset
- Skip 1 pad frame in readback (101 words = 404 bytes)
- Mask ECC word (word #50 per frame) — volatile on readback
- Track BRAM content frames separately — volatile after startup
- Report logic vs BRAM vs ECC mismatches independently
- PASS if logic mismatches < 1% (without .rbd/.msd, some volatile bits expected)

Also corrects 5 stale experience entries (FPGA-001/006/007/008/009) that
incorrectly blamed CPLD 0xFFFE as dead hardware — real bug was bitrev().

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gHashTag pushed a commit that referenced this pull request Mar 15, 2026
@claude
feat(fpga): DLC-10 JTAG switcher — bitrev fix + UG470 verify + tri CLI
6754b73 
Cherry-picked 3 fixes from feat/fpga-dlc10-breakthrough:
- bitrev() in CFG_OUT reads (STAT 0x02089E3F → 0x401079FC, DONE=YES)
- bitrev() in readback (raw TDO byte reversal)
- UG470-compliant verify: pad frame skip, ECC word #50 mask, BRAM tracking

Added `tri fpga jtag` CLI wrapper (write/readback/verify/status/idcode/dna/reg/probe/debug).

All 9 JTAG commands PASS on XC7A100T via DLC-10 Platform Cable USB II.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gHashTag pushed a commit that referenced this pull request Mar 15, 2026
@claude
fix(fpga): UG470-compliant verify — frame alignment, ECC/BRAM masking
7da2fde 
Replaced naive byte-by-byte comparison with proper UG470 verify:
- Parse FDRI write command to find frame data offset
- Skip pad frame in readback (101 words)
- Mask ECC word #50 per frame (volatile on readback)
- Track BRAM content frames separately (volatile after startup)
- Report logic vs BRAM vs ECC mismatches independently
- PASS if logic mismatches < 1% (without .rbd/.msd mask files)

Results on hslm_full_top.bit: 3753 frames compared, 2106 clean,
ECC=3021 (expected), BRAM=1768 (expected), logic=61728 (4% — needs
FDRI multi-range parsing for full accuracy, STAT CRC=OK confirms
correct programming).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Mar 18, 2026
@gHashTag @claude
feat(hybrid): IglaHybridChat v2.1 — heap TVC + live provider health +…
057effe 
… canvas wave [Golden Chain #50]

- TVCCorpus.initHeap(allocator): eliminates 2.15 GB stack frame, Debug builds work
- Provider health wired to llmCascade(): recordSuccess/recordFailure with real latency
- Circuit breaker: skip providers with 3+ consecutive failures
- Canvas wave ring modulated by g_last_wave_state (source hue, confidence pulse, learning glow)
- 2947/2954 tests pass (+4 new v2.1 tests)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Mar 18, 2026
@gHashTag @claude
feat(agent): Ralph sleep-wake daemon — native Zig, GitHub polling, Cl…
70f2df2 
…aude CLI spawn (#50)

8 Zig modules (~500 LOC):
- main.zig: entry point, env vars, CLI args
- agent_loop.zig: WAKE→WORK→SLEEP cycle
- identity.zig: load .ralph/IDENTITY.md
- context_builder.zig: assemble structured prompt
- github_poller.zig: fetch pending issues via HTTP
- claude_runner.zig: spawn `claude` CLI as child process
- handover.zig: read/write .ralph/HANDOVER.md
- state.zig: file-based key-value persistence

Build: `zig build agent` / `zig build agent -- --single-shot`

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Antigravity Agent <antigravity@vibee.org>
Co-authored-by: Claude <noreply@anthropic.com>
gHashTag pushed a commit that referenced this pull request Mar 18, 2026
@claude
fix(fpga): UG470-compliant verify with frame alignment and ECC/BRAM m…
232915a 
…asking

cmd_verify rewritten to properly compare readback vs .bit file:
- Find FDRI write command in .bit → extract frame data at correct offset
- Skip 1 pad frame in readback (101 words = 404 bytes)
- Mask ECC word (word #50 per frame) — volatile on readback
- Track BRAM content frames separately — volatile after startup
- Report logic vs BRAM vs ECC mismatches independently
- PASS if logic mismatches < 1% (without .rbd/.msd, some volatile bits expected)

Also corrects 5 stale experience entries (FPGA-001/006/007/008/009) that
incorrectly blamed CPLD 0xFFFE as dead hardware — real bug was bitrev().

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gHashTag pushed a commit that referenced this pull request Mar 18, 2026
@claude
feat(fpga): DLC-10 JTAG switcher — bitrev fix + UG470 verify + tri CLI
8b8afba 
Cherry-picked 3 fixes from feat/fpga-dlc10-breakthrough:
- bitrev() in CFG_OUT reads (STAT 0x02089E3F → 0x401079FC, DONE=YES)
- bitrev() in readback (raw TDO byte reversal)
- UG470-compliant verify: pad frame skip, ECC word #50 mask, BRAM tracking

Added `tri fpga jtag` CLI wrapper (write/readback/verify/status/idcode/dna/reg/probe/debug).

All 9 JTAG commands PASS on XC7A100T via DLC-10 Platform Cable USB II.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gHashTag pushed a commit that referenced this pull request Mar 18, 2026
@claude
fix(fpga): UG470-compliant verify — frame alignment, ECC/BRAM masking
a723bdd 
Replaced naive byte-by-byte comparison with proper UG470 verify:
- Parse FDRI write command to find frame data offset
- Skip pad frame in readback (101 words)
- Mask ECC word #50 per frame (volatile on readback)
- Track BRAM content frames separately (volatile after startup)
- Report logic vs BRAM vs ECC mismatches independently
- PASS if logic mismatches < 1% (without .rbd/.msd mask files)

Results on hslm_full_top.bit: 3753 frames compared, 2106 clean,
ECC=3021 (expected), BRAM=1768 (expected), logic=61728 (4% — needs
FDRI multi-range parsing for full accuracy, STAT CRC=OK confirms
correct programming).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@gHashTag gHashTag mentioned this pull request Apr 29, 2026
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@gHashTag