Target adapters

Each target adapter translates the wep harness into the format the agent understands. The neutral source of truth (.wep/) is shared across targets. The layouts below describe a full installation; --modules can install a smaller set. See modules.

Codex

AGENTS.md                                ← main instructions (protected)
.agents/skills/wep-init/SKILL.md
.agents/skills/wep-doctor/SKILL.md
.agents/skills/wep-task/SKILL.md
.agents/skills/wep-review/SKILL.md
.agents/skills/wep-workflow/SKILL.md
.agents/skills/wep-write-policy/SKILL.md
.codex/agents/wep-judge.toml          ← enabled by normal init; also managed by `wep judge enable`
.codex/agents/wep-comprehension.toml  ← opt-in: `wep comprehension enable`
.codex/agents/wep-implementer.toml    ← opt-in: `wep impl enable`

AGENTS.md is the primary instruction file Codex reads. Skills are loaded on demand when the agent invokes a workflow. Judge uses Codex’s native subagent system (.codex/agents/*.toml).

---

Claude Code

CLAUDE.md                                ← main instructions (protected)
.claude/commands/wep-doctor.md
.claude/commands/wep-task.md
.claude/skills/wep-init/SKILL.md
.claude/skills/wep-doctor/SKILL.md
.claude/skills/wep-task/SKILL.md
.claude/skills/wep-review/SKILL.md
.claude/skills/wep-workflow/SKILL.md
.claude/skills/wep-write-policy/SKILL.md
.claude/agents/wep-judge.md           ← enabled by normal init; also managed by `wep judge enable`
.claude/agents/wep-comprehension.md   ← opt-in: `wep comprehension enable`
.claude/agents/wep-implementer.md     ← opt-in: `wep impl enable`

CLAUDE.md is the primary instruction file Claude Code reads. Commands are slash-command shortcuts. Skills load on demand. Judge uses Claude Code’s native subagent system (.claude/agents/*.md with YAML frontmatter). Enabling the judge also installs a guard hook in .claude/settings.local.json to refuse file-writing tools while a judge validation run holds its lock. Other hosts rely on the after-run change check.

---

GitHub Copilot

.github/copilot-instructions.md          ← main instructions (protected)
.github/instructions/wep.instructions.md
.github/prompts/wep-doctor.prompt.md
.github/prompts/wep-task.prompt.md
.github/prompts/wep-workflow.prompt.md
.github/prompts/wep-write-policy.prompt.md
.github/skills/wep-init/SKILL.md
.github/skills/wep-doctor/SKILL.md
.github/skills/wep-task/SKILL.md
.github/skills/wep-review/SKILL.md
.github/skills/wep-workflow/SKILL.md
.github/skills/wep-write-policy/SKILL.md
.github/agents/wep-judge.agent.md     ← enabled by normal init; also managed by `wep judge enable`
.github/agents/wep-comprehension.agent.md ← opt-in: `wep comprehension enable`
.github/agents/wep-implementer.agent.md   ← opt-in: `wep impl enable`

Copilot reads copilot-instructions.md globally. Instructions files scope rules to file patterns. Prompts are reusable conversation starters. Judge uses Copilot’s native agent system (.github/agents/*.agent.md).

---

Pi

.pi/prompts/wep-doctor.md
.pi/prompts/wep-task.md
.pi/prompts/wep-workflow.md
.pi/prompts/wep-write-policy.md
.pi/skills/wep-init/SKILL.md
.pi/skills/wep-doctor/SKILL.md
.pi/skills/wep-task/SKILL.md
.pi/skills/wep-review/SKILL.md
.pi/skills/wep-workflow/SKILL.md
.pi/skills/wep-write-policy/SKILL.md

Wepwawet has no native subagent adapter for Pi. Judge, comprehension, and implementer agents are not available for Pi. Their enable commands skip Pi automatically; workflow prompts and skills remain available.

---

Neutral source of truth (all targets)

.wep/
  config.json          ← modules, defaults, installed version, agent settings
  manifest.json        ← hashes of generated files written by wep
  policy.json          ← merge strategy, blocked read patterns, context budget
  local/               ← Git-ignored review, implementation, tracing, and learning records
  judge/               ← review contract and schema
  comprehension/       ← comprehension contract and schemas
  wiki/
    index.md           ← wiki catalog
    overview.md        ← project name, installed targets, quick reference
    architecture.md
    conventions.md
    testing.md
    workflows.md
    decisions.md
    agent-setup.md     ← written by `wep doctor`
    gaps.md            ← regenerated by `wep doctor`
    recommendations.md ← regenerated by `wep doctor`
    instruction-audit.md ← persistent semantic review written by the Doctor skill
    write-policy.md
    log.md
  tasks/
    _template/         ← capsule template files
    TASK-XXX-<slug>/   ← one directory per task
  targets/
    codex.md
    claude.md
    copilot.md
    pi.md

The selected modules determine which neutral files are written; a full init writes the layout above regardless of target. Target adapters are thin — they translate the harness into the format each agent expects.