|
13 | 13 | - Minimal markdown markers, no unnecessary formatting, minimal emojis. |
14 | 14 | - Reference issue numbers in the format `#<issue-number>` for easy linking. |
15 | 15 |
|
| 16 | +# [2026-04-28] Dev Log: agent-context consolidation — ~/.claude as single source of truth [WIP] |
| 17 | +- Why: agent customizations are scattered across three directories (=~/.claude/=, =~/.codex/=, =~/.agents/=), 10 concepts are duplicated between Claude Code agents and Codex skills and are already diverging, and the shared global instructions (GENERAL-AGENTS.md) copied into both =~/.claude/CLAUDE.md= and =~/.codex/AGENTS.md= are 30 lines out of sync |
| 18 | +- What: |
| 19 | + - **[PLANNING]** Phase 1 — establish source of truth: =~/.claude/= is canonical; Claude Code agents/ are authoritative for the 10 duplicated concepts; =~/.claude/CLAUDE.md= is the master global instructions file |
| 20 | + - **[PLANNING]** Phase 2 — Codex sync mechanism: decision needed between symlinking =~/.codex/AGENTS.md= -> =~/.claude/CLAUDE.md= vs. a sync script that copies with Codex-specific adaptations; leaning toward symlink for zero-drift maintenance |
| 21 | + - **[PLANNING]** Phase 3 — deduplicate Codex skills: remove from =~/.codex/skills/= all 10 Claude Code agent duplicates (linus-committer, linus-rants, comment-flusher, etc.) plus 8 skills already covered by the superpowers plugin in CC (brainstorming, executing-plans, writing-plans, dispatching-parallel-agents, subagent-driven-development, skill-creator-6rok, using-superpowers, codex-primary-runtime); retain Codex-only skills: docker-compose-safe-deploy, systematic-debugging, test-driven-development, verification-before-completion, finishing-a-development-branch, using-git-worktrees, receiving-code-review, writing-skills |
| 22 | + - **[PLANNING]** Phase 4 — clean =~/.agents/=: this directory is dead weight; Claude Code does not scan it; Codex only scans .agents/ walking up from cwd, not from HOME; decide whether to archive the telegram TS package and repurpose the dir as a shared skill source, or delete it entirely |
| 23 | + - **[PLANNING]** Phase 5 — purge stale files: delete all .bak/.backup files from =~/.claude/= and =~/.codex/= (~1MB), delete empty =~/.codex/instructions.md=, clean =~/.claude/.claude.json= backup chain |
| 24 | + - **[PLANNING]** Phase 6 — document the layout: brief README or section in CLAUDE.md covering what lives where, how to add new skills/agents, and the Codex sync workflow |
| 25 | + |
| 26 | +- Open decisions: |
| 27 | + 1. Symlink =AGENTS.md= -> =CLAUDE.md=, or keep separate with a sync script? (symlink eliminates drift but loses ability to add Codex-specific front matter) |
| 28 | + 2. Kill the 10 Codex duplicates entirely, or keep for standalone Codex use without deva? (Claude Code + superpowers already covers the gap when running inside deva) |
| 29 | + 3. =~/.agents/=: repurpose as shared skill source readable by both tools, or delete? |
| 30 | + 4. Should unique Codex workflow skills (TDD, systematic-debugging, finishing-a-development-branch, etc.) be ported back into Claude Code agents? |
| 31 | + |
| 32 | +- Result: **[WIP — no files changed yet]** planning entry only; execution blocked on the four decisions above |
| 33 | + |
16 | 34 | # [2026-04-13] Dev Log: adopt smux tmux-bridge as layer-2 agent comms |
17 | 35 | - Why: deva-v2 proposal left the agent-to-agent transport CLI as an open question; smux (github.com/ShawnPana/smux) shipped exactly the shape we sketched (read/type/keys/label/envelope) in 403 lines of bash, MIT-licensed. Reinventing it would be waste. The existing `deva-bridge-tmux` (socat TCP) is layer 1 (kernel boundary); `tmux-bridge` is layer 2 (semantic). They compose cleanly. |
18 | 36 | - What: |
|
0 commit comments