Files

gsxdsm 0f81a83800 fix(core): sync default-triage prompt with engine's TRIAGE_SYSTEM_PROMPT

Root cause of "fn_review_spec was never called" tracked through the diagnostic
chain: the prompt actually sent to triage agents has zero `fn_*` tokens —
because resolveAgentPrompt("triage", ...) returns the BUILTIN_AGENT_PROMPTS
default-triage template (TRIAGE_PROMPT_TEXT in core/agent-prompts.ts), and that
template was forked from an older version that never had the "MUST call
fn_review_spec()" workflow nor any fn_-prefixed tool names. The fallback
`|| TRIAGE_SYSTEM_PROMPT` in engine/triage.ts never fires because the core
template is non-empty.

So the model writes PROMPT.md, doesn't see any instruction to review it, and
ends. zai/glm-5.1 happened to call fn_review_spec from training-pattern
inertia; Sonnet via pi-claude-cli stopped at write — same prompt, same bug.

Replace TRIAGE_PROMPT_TEXT with the engine's up-to-date TRIAGE_SYSTEM_PROMPT
verbatim (fn_-prefixed tools, fn_review_spec workflow, subtask breakdown,
project-commands handling, frontend UX criteria injection). Also remove the
diagnostic-only console.error lines added during this debugging session — the
core fix is now elsewhere and the noise isn't worth keeping.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-04-25 17:24:03 -07:00

src

fix(core): sync default-triage prompt with engine's TRIAGE_SYSTEM_PROMPT

2026-04-25 17:24:03 -07:00

CHANGELOG.md

chore(release): v0.2.7

2026-04-25 11:38:48 -07:00

index.ts

fix(pi-claude-cli): source MCP tool list from session Context, not pi.getAllTools

2026-04-25 14:31:15 -07:00

LICENSE

feat: add "Anthropic — via Claude CLI" as a first-class provider

2026-04-23 23:02:34 -07:00

package.json

chore(release): v0.2.7

2026-04-25 11:38:48 -07:00

README.md

feat: add "Anthropic — via Claude CLI" as a first-class provider

2026-04-23 23:02:34 -07:00

tsconfig.json

chore: consolidate test files into __tests__/ dirs and clean stray engine artifacts

2026-04-24 19:39:20 -07:00

UPSTREAM.md

feat: add "Anthropic — via Claude CLI" as a first-class provider

2026-04-23 23:02:34 -07:00

vitest.config.ts

feat: add "Anthropic — via Claude CLI" as a first-class provider

2026-04-23 23:02:34 -07:00

README.md

pi-claude-cli

A pi extension that routes LLM calls through the Claude Code CLI as a subprocess. Use your Claude Pro/Max subscription as the LLM backend — no API key, no separate billing.

How it works

The extension registers as a custom pi provider exposing all Claude models. Each request spawns a claude -p subprocess using the stream-json wire protocol, with --resume on follow-up turns to reuse the CLI's session state instead of replaying full history. Claude proposes tool calls, pi executes them natively. Custom pi tools are exposed to Claude via a schema-only MCP server.

Requirements

Claude Code CLI installed and authenticated (claude on PATH)
A Claude Pro or Max subscription
pi or GSD

Installation

Add to ~/.gsd/agent/settings.json:

{
  "packages": ["npm:pi-claude-cli"]
}

Then select a Claude model via /model in the interactive UI. All Claude models appear under the pi-claude-cli provider.

Features

Streams text, thinking, and tool call tokens in real-time
Maps tool names and arguments bidirectionally between Claude and pi
Exposes custom pi tools to Claude via MCP (schema-only, no execution)
Break-early pattern prevents Claude CLI from auto-executing tools
Session resume via --resume eliminates history replay on follow-up turns
Configurable thinking effort with elevated budgets for Opus models
Cross-platform subprocess management (Windows, macOS, Linux)
Inactivity timeout and process registry for cleanup

License

MIT