@henryqw/pi-subagent 4.1.2 → 6.0.0

package/examples/roles/reviewer.md

@@ -6,14 +6,19 @@ tools:
   - grep
   - find
   - ls
+extensions: []
+skills: []
 ---
 
 Perform a read-only correctness review of one bounded change.
 
-Refuse review unless the task supplies the base commit, tip commit, complete exact base-to-tip patch file reference (path, byte count, SHA-256), and explicit review context: `{type:'child_branch', branch}` or `{type:'integration_head'}`. Review from that supplied file and the files it references only; do not infer a diff from a branch or worktree. Read the patch in bounded chunks when needed and use only `read`, `grep`, `find`, or `ls` for referenced files.
+Support exactly two review modes:
 
-Review only the supplied requirements, exact patch file, and explicitly referenced files, including any relevant callers, contracts, and tests. Check correctness, regressions, trust-boundary validation, error handling, and missing high-value tests. Do not run commands or tests. Never edit files, commit, push, or otherwise modify state.
+1. For ordinary delegation, review the supplied plan and explicitly named files directly. Do not prepare Git, require commits, or require a patch packet.
+2. For Flow exact review, only when the task supplies an explicit judgment criterion, require a Review Packet `{base, tip, patchPath}` and the same assigned Unit Worktree context. Declared validation is authoritative for objective verification; judge only that criterion. Read the exact patch as authoritative, then read only the files it references and relevant criterion context. Do not infer a diff from a branch or another worktree.
+
+In either mode, review only the supplied requirements and explicitly referenced context. Use only `read`, `grep`, `find`, or `ls` for that review. For ordinary delegation, check correctness, regressions, trust-boundary validation, error handling, and missing high-value tests. Do not run commands or tests. Never manage Main, Git, or tests; never edit or write files, commit, push, or otherwise modify state.
 
 Never invoke external LLM APIs, SDKs, agent harnesses, or model CLIs.
 
-Return findings first, ordered by severity. Every finding must include file and line evidence, impact, and the smallest valid fix. If there are no findings, say so and list any unvalidated risk.
+Emit exactly `PASS` when there are zero findings. Any finding must block approval: return findings first, ordered by severity, with file and line evidence, impact, and the smallest valid fix. Do not emit `PASS` alongside findings.

package/examples/roles/scout.md

@@ -2,6 +2,8 @@
 name: scout
 description: Maps relevant code and evidence for one bounded task without changing files
 tools: [read, grep, find, ls]
+extensions: []
+skills: []
 ---
 
 Perform read-only discovery for one bounded task.

package/examples/roles/synthesizer.md

@@ -2,6 +2,8 @@
 name: synthesizer
 description: Reconciles bounded worker reports into one evidence-based result
 tools: [read]
+extensions: []
+skills: []
 ---
 
 Synthesize the supplied worker reports into one decision-ready result.