@leing2021/super-pi 0.23.3 → 0.23.5

package/README.md

@@ -89,7 +89,7 @@ New conversation overhead: **~2,600 tokens** (1.3% of 200K context).
 
 Progressive loading: only needed skills loaded on-demand.
 
-Full evaluation → [`docs/token-cost-evaluation.md`](docs/token-cost-evaluation.md)
+Full evaluation → see Token Cost section above.
 
 ---
 

package/extensions/ce-core/index.ts

@@ -332,6 +332,7 @@ const contextHandoffParams = Type.Object({
   openDecisions: Type.Optional(Type.Array(Type.String(), { description: "Pending decisions that affect next steps" })),
   recentlyAccessedFiles: Type.Optional(Type.Array(Type.String(), { description: "Files recently read or edited (defaults to activeFiles)" })),
   compressionRisk: Type.Optional(Type.Array(Type.String(), { description: "Context compression risks to watch for" })),
+  activeRules: Type.Optional(Type.Array(Type.String(), { description: "1-5 must-know rules for continuation (TDD gates, constraints, do-not-repeat)" })),
 })
 
 const patternExtractorParams = Type.Object({
@@ -772,6 +773,7 @@ export default function ceCoreExtension(pi: ExtensionAPI) {
         openDecisions: params.openDecisions,
         recentlyAccessedFiles: params.recentlyAccessedFiles,
         compressionRisk: params.compressionRisk,
+        activeRules: params.activeRules,
       })
 
       return {
@@ -891,4 +893,4 @@ export {
 export { normalizeSlug } from "./utils/name-utils"
 export { filterBashOutput } from "./tools/bash-output-filter"
 export { filterReadOutput } from "./tools/read-output-filter"
-export { COMPACTION_FOCUS_INSTRUCTIONS, TURN_PREFIX_FOCUS_INSTRUCTIONS } from "./tools/compaction-optimizer"
+export { COMPACTION_FOCUS_INSTRUCTIONS } from "./tools/compaction-optimizer"

package/extensions/ce-core/tools/compaction-optimizer.ts

@@ -27,12 +27,4 @@ export const COMPACTION_FOCUS_INSTRUCTIONS = `Additional focus for this summary:
 6. If any tests were run, summarize results by: file, pass/fail count, and specific failure messages
 7. Note any blocked items and their exact error state`
 
-/**
- * Additional instructions for turn-prefix summaries (split turns).
- * These are more concise since the turn suffix is retained.
- */
-export const TURN_PREFIX_FOCUS_INSTRUCTIONS = `Focus the turn-prefix summary on:
-- What the user originally asked for
-- Key decisions made in the prefix
-- Exact file paths and identifiers needed to understand the retained suffix
-Skip reasoning details — only keep actionable context.`
+

package/extensions/ce-core/tools/context-handoff.ts

@@ -37,6 +37,7 @@ export interface ContextHandoffInput {
   openDecisions?: string[]
   recentlyAccessedFiles?: string[]
   compressionRisk?: string[]
+  activeRules?: string[]
 }
 
 export interface ContextStateEntry {
@@ -54,6 +55,7 @@ export interface ContextStateEntry {
   openDecisions: string[]
   recentlyAccessedFiles: string[]
   compressionRisk: string[]
+  activeRules: string[]
   recommendNewSession: boolean
   updatedAt: string
 }
@@ -77,6 +79,7 @@ export interface ContextHandoffResult {
   openDecisions?: string[]
   recentlyAccessedFiles?: string[]
   compressionRisk?: string[]
+  activeRules?: string[]
   updatedAt?: string
   // Validation fields
   ok?: boolean
@@ -154,6 +157,7 @@ function buildDefaultHandoffMarkdown(input: {
   openDecisions: string[]
   recentlyAccessedFiles: string[]
   compressionRisk: string[]
+  activeRules: string[]
 }): string {
   const currentTask = input.nextStage
     ? `Continue from ${input.currentStage} to ${input.nextStage}.`
@@ -194,6 +198,9 @@ function buildDefaultHandoffMarkdown(input: {
     "## Active Files",
     activeFiles,
     "",
+    "## Active Rules",
+    formatBullets(input.activeRules),
+    "",
     "## Recently Accessed Files",
     formatBullets(input.recentlyAccessedFiles),
     "",
@@ -245,6 +252,7 @@ function normalizeStateEntry(raw: unknown): ContextStateEntry | null {
       ? toStringArray(state.recentlyAccessedFiles)
       : activeFiles.slice(0, 5),
     compressionRisk: toStringArray(state.compressionRisk),
+    activeRules: toStringArray(state.activeRules),
     recommendNewSession: typeof state.recommendNewSession === "boolean" ? state.recommendNewSession : false,
     updatedAt: typeof state.updatedAt === "string" ? state.updatedAt : new Date(0).toISOString(),
   }
@@ -314,6 +322,7 @@ async function save(input: ContextHandoffInput): Promise<ContextHandoffResult> {
     ? input.recentlyAccessedFiles
     : activeFiles.slice(0, 5)
   const compressionRisk = input.compressionRisk ?? []
+  const activeRules = input.activeRules ?? []
 
   const handoffMarkdown = input.handoffMarkdown?.trim().length
     ? input.handoffMarkdown
@@ -329,6 +338,7 @@ async function save(input: ContextHandoffInput): Promise<ContextHandoffResult> {
       openDecisions,
       recentlyAccessedFiles,
       compressionRisk,
+      activeRules,
     })
 
   const recommendNewSession = computeRecommendNewSession(currentStage, nextStage, contextHealth)
@@ -356,6 +366,7 @@ async function save(input: ContextHandoffInput): Promise<ContextHandoffResult> {
     openDecisions,
     recentlyAccessedFiles,
     compressionRisk,
+    activeRules,
     recommendNewSession,
     updatedAt: new Date().toISOString(),
   }
@@ -379,6 +390,7 @@ async function save(input: ContextHandoffInput): Promise<ContextHandoffResult> {
     openDecisions,
     recentlyAccessedFiles,
     compressionRisk,
+    activeRules,
     recommendNewSession,
     updatedAt: state.updatedAt,
   }
@@ -419,6 +431,7 @@ async function load(input: ContextHandoffInput): Promise<ContextHandoffResult> {
     openDecisions: state.openDecisions,
     recentlyAccessedFiles: state.recentlyAccessedFiles,
     compressionRisk: state.compressionRisk,
+    activeRules: state.activeRules,
     recommendNewSession: state.recommendNewSession,
     handoffMarkdown: markdown,
     updatedAt: state.updatedAt,
@@ -453,6 +466,7 @@ async function latest(input: ContextHandoffInput): Promise<ContextHandoffResult>
     openDecisions: state.openDecisions,
     recentlyAccessedFiles: state.recentlyAccessedFiles,
     compressionRisk: state.compressionRisk,
+    activeRules: state.activeRules,
     recommendNewSession: state.recommendNewSession,
     updatedAt: state.updatedAt,
   }
@@ -755,6 +769,7 @@ async function status(input: ContextHandoffInput): Promise<ContextHandoffResult>
     openDecisions: state.openDecisions,
     recentlyAccessedFiles: state.recentlyAccessedFiles,
     compressionRisk: state.compressionRisk,
+    activeRules: state.activeRules,
     recommendNewSession: state.recommendNewSession,
     updatedAt: state.updatedAt,
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@leing2021/super-pi",
-  "version": "0.23.3",
+  "version": "0.23.5",
   "private": false,
   "type": "module",
   "description": "Pi-native Compound Engineering package for iterative development workflows",

package/rules/common/development-workflow.md

@@ -13,6 +13,7 @@ The Feature Implementation Workflow describes the development pipeline: research
    - **Check package registries:** Search npm, PyPI, crates.io, and other registries before writing utility code. Prefer battle-tested libraries over hand-rolled solutions.
    - **Search for adaptable implementations:** Look for open-source projects that solve 80%+ of the problem and can be forked, ported, or wrapped.
    - Prefer adopting or porting a proven approach over writing net-new code when it meets the requirement.
+   - **Source-driven trigger:** When implementation depends on a framework/library API, version-specific behavior, or a recommended pattern, verify against official documentation and cite key sources in the output. Pure logic, renaming, or in-project pattern reuse does not require external citation.
 
 1. **Plan First**
    - Use **planner** agent to create implementation plan

package/skills/01-brainstorm/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 01-brainstorm
-description: "Brainstorm requirements with three modes: CE discovery, Startup Diagnostic, Builder Mode."
+description: "Discover requirements through structured multi-round dialog. Use when the request is ambiguous, needs discovery, or describes a new idea/product."
 ---
 
 # Brainstorm

package/skills/02-plan/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 02-plan
-description: "Turn requirements into a plan. Optional CEO-style strategic review after."
+description: "Turn requirements into an execution-ready plan with TDD-gated implementation units. Use when a brainstorm artifact exists and is ready for planning."
 ---
 
 # Plan
@@ -39,13 +39,15 @@ Every unit follows **RED → GREEN → REFACTOR**:
 
 ## Planning flow
 
-1. Read relevant brainstorm from `docs/brainstorms/`
-2. Run solution search (keywords → grep frontmatter → read top 3)
-3. Gather repository context
-4. If plan exists: use `plan_diff` `compare` → review with user → `patch`
-5. If no plan: write new plan under `docs/plans/` using `references/plan-template.md`
-6. Structure work using `references/implementation-unit-template.md`
-7. Verify every unit follows TDD gates
+1. **Load context**: consume latest handoff before any broad file reads — `context_handoff load` or read `.context/compound-engineering/handoffs/latest.md`. If found, use `activeFiles` and `blocker` as starting point. If not found, proceed normally (new project).
+2. Read relevant brainstorm from `docs/brainstorms/`
+3. Run solution search (keywords → grep frontmatter → read top 3)
+4. Gather repository context
+5. **Source-driven check:** For each unit that involves framework/library APIs, add a note: "Verify against official docs before implementing."
+6. If plan exists: use `plan_diff` `compare` → review with user → `patch`
+6. If no plan: write new plan under `docs/plans/` using `references/plan-template.md`
+7. Structure work using `references/implementation-unit-template.md`
+8. Verify every unit follows TDD gates
 
 ## Optional: CEO Review
 

package/skills/03-work/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 03-work
-description: Execute plan-driven work in a controlled Phase 1 workflow.
+description: "Execute plan units with parallel subagents, TDD enforcement, and checkpoint resume. Use when a plan path is ready for implementation."
 ---
 
 # Work
@@ -39,18 +39,39 @@ Every step follows **RED → GREEN → REFACTOR**:
 - Missing evidence test passed after implementation
 - Tests added only after code
 
+## Stop-the-line rule (Hard gate)
+
+When any unexpected failure occurs during execution:
+
+1. **STOP** adding features or making changes
+2. **PRESERVE** evidence (error output, repro steps)
+3. **DIAGNOSE** root cause — reproduce, localize, reduce
+4. **FIX** the root cause, not the symptom
+5. **GUARD** with a regression test
+6. **RESUME** only after verification passes
+
+Anti-rationalization — when a gate fails or evidence is missing:
+- Do not rationalize, downgrade, or explain away the failure.
+- Stop, report the blocker with evidence, and either fix the root cause or ask for direction.
+- Do not continue unrelated implementation after failed verification.
+
+This is a hard gate — do not push past a failing test or broken build to continue implementation. Errors compound.
+
 ## Workflow
 
-1. Detect input type (plan path vs bare prompt)
-2. Read implementation units if plan path
-3. Load `session_checkpoint` to skip completed units
-4. Use `task_splitter` for dependency analysis
-5. Execute: **inline mode** by default, `ce_parallel_subagent` for independent units
-6. Follow TDD per unit: RED → minimal code → GREEN → refactor → unit-level **verification**
-7. Record progress via `references/progress-update-format.md`
-8. Save `session_checkpoint` after each unit
-9. On failure: `session_checkpoint` `fail` → `retry` → follow strategy
-10. Provide completion report (see `references/completion-report.md`)
-11. Handoff to `04-review` using `references/handoff.md`
+1. **Load context**: consume latest handoff before any broad file reads — `context_handoff load` or read `.context/compound-engineering/handoffs/latest.md`. If found, use `activeFiles`, `blocker`, `verification`, `activeRules` as starting point. If not found, proceed normally.
+2. Detect input type (plan path vs bare prompt)
+3. Read implementation units if plan path
+4. Load `session_checkpoint` to skip completed units
+5. Use `task_splitter` for dependency analysis
+6. Execute: **inline mode** by default, `ce_parallel_subagent` for independent units
+7. Follow TDD per unit: RED → minimal code → GREEN → refactor → unit-level **verification**
+8. **Source-driven gate:** Before implementing framework/library-specific code, verify the API or pattern against official documentation. Flag unverified patterns as UNVERIFIED in output.
+9. Record progress via `references/progress-update-format.md`
+9. Save `session_checkpoint` after each unit
+10. On failure: `session_checkpoint` `fail` → `retry` → follow strategy
+11. Provide completion report (see `references/completion-report.md`)
+12. **Save handoff**: `context_handoff save` with current stage, next stage, activeFiles, blocker, verification, activeRules
+13. Handoff to `04-review` using `references/handoff.md`
 
 Before finishing this skill, apply the completion checklist in [shared pipeline instructions](../references/pipeline-config.md).

package/skills/04-review/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 04-review
-description: "Review code with structured findings. Optional browser QA and regression tests."
+description: "Review code changes across five axes with evidence-first findings. Use after implementation is complete and before committing."
 ---
 
 # Review
@@ -46,14 +46,15 @@ Code review is **technical evaluation**, not social performance:
 
 ## Workflow
 
-1. Determine diff scope from branch or explicit target
-2. Collect stats (files, insertions, deletions) → call `review_router`
-3. Read matching plan artifact
-4. Run solution search
-5. Apply each reviewer persona from `review_router`
-6. Merge into structured findings
-7. Verify each finding against codebase
-8. Apply autofixes, re-run tests, re-review if needed
+1. **Load context**: consume latest handoff before any broad file reads — `context_handoff load` or read `.context/compound-engineering/handoffs/latest.md`. If found, use `activeFiles`, `artifacts.plan` as starting point. If not found, proceed normally.
+2. Determine diff scope from branch or explicit target
+3. Collect stats (files, insertions, deletions) → call `review_router`
+4. Read matching plan artifact
+5. Run solution search
+6. Apply each reviewer persona from `review_router`
+7. Merge into structured findings
+8. Verify each finding against codebase
+9. Apply autofixes, re-run tests, re-review if needed
 
 ## Optional: QA Test Mode
 

package/skills/04-review/references/reviewer-selection.md

@@ -10,6 +10,8 @@ Use the `review_router` tool to determine which reviewers to apply based on diff
 
 ## Reviewer personas
 
+All reviewers evaluate changes across five axes: correctness, readability, architecture, security, performance. Base reviewers cover axes 1–3 by default; conditional reviewers add depth on specific axes.
+
 ### Base reviewers (always active)
 
 - **correctness-reviewer**: Logical correctness, intended behavior, edge cases.
@@ -19,6 +21,6 @@ Use the `review_router` tool to determine which reviewers to apply based on diff
 ### Conditional reviewers (routed by `review_router`)
 
 - **security-reviewer**: Triggered when auth, permissions, tokens, sessions, or crypto files change. Reviews for injection, auth bypass, credential leakage.
-- **performan04-reviewer**: Triggered when query, cache, database, or streaming files change. Reviews for N+1, unnecessary allocation, missing indexes.
+- **performance-reviewer**: Triggered when query, cache, database, or streaming files change. Reviews for N+1, unnecessary allocation, missing indexes.
 - **integration-reviewer**: Triggered when CI/CD, Docker, package.json, or config files change. Reviews for dependency conflicts, build breakage, deployment issues.
 - **thoroughness-reviewer**: Triggered for large diffs (5+ files or 300+ lines). Reviews for incomplete refactors, missed callers, inconsistent changes.

package/skills/05-learn/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 05-learn
-description: Capture solved problems as reusable solution artifacts.
+description: "Capture solved problems as searchable solution artifacts. Use after a workflow loop completes or a non-trivial problem is solved."
 ---
 
 # Learn

package/skills/06-next/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 06-next
-description: "Inspect workflow state and recommend the single best next Compound Engineering skill. Use --verbose for a full status report."
+description: "Inspect workflow artifacts and recommend the single best next skill. Use when unsure what to run next."
 ---
 
 # Next
@@ -22,8 +22,9 @@ Use this skill when the user wants to know what to run next in the Compound Engi
 When the user asks "what should I do next?", "continue", or runs `/skill:06-next`:
 
 1. Call `workflow_state` with the repo root
-2. Apply recommendation logic from `references/recommendation-logic.md`
-3. Return: skill name, reason (1-2 lines), brief workflow state summary
+2. Inspect `workflow_state.context` first — apply **context-first priority chain** from `references/recommendation-logic.md` (health → blocker → recommendNewSession → nextStage → mismatch → fallback)
+3. If no context signals trigger, fall back to artifact-count rules
+4. Return: skill name, reason (1-2 lines), brief workflow state summary
 
 ### Verbose mode: "full status"
 
@@ -31,7 +32,8 @@ When the user asks "show status", "what's the current state", or uses `--verbose
 
 1. Call `workflow_state` with the repo root
 2. Call `session_history` with `latest` operation
-3. Return: latest artifacts (path + summary), status of each phase, recommended next step
+3. Include context health assessment from `workflow_state.context` in the status report
+4. Return: latest artifacts (path + summary), status of each phase, context health, recommended next step
 
 ## Artifact locations
 

package/skills/06-next/references/recommendation-logic.md

@@ -1,6 +1,51 @@
 # Recommendation logic
 
-Apply these rules in order. Return the first match.
+Apply these rules in strict priority order. Return the first match.
+
+## Context-first priority chain
+
+`06-next` must call `workflow_state` and inspect `workflow_state.context` **before** applying artifact-count rules. Context health and runtime state override stale artifact counts.
+
+Field mapping: the requirement uses `context.health` as a conceptual alias. Use the actual field name `workflow_state.context.contextHealth`.
+
+### Priority 1: Critical context health
+
+If `context.contextHealth` is `"critical"`:
+- Recommend: save a `context_handoff` with full handoff-lite template, then open a **new session**.
+- Reason: Current session context is critically inflated. Continuing will degrade decision quality and waste tokens.
+- Output: handoff-save guidance + copyable new-session prompt with repo path and artifact references.
+
+### Priority 2: Active blocker
+
+If `context.blocker` exists and is not `"N/A"` or a placeholder:
+- Recommend: return to the current stage (`context.currentStage`) and resolve the blocker before advancing.
+- Reason: A known blocker exists in the current pipeline stage.
+- Output: `/skill:<currentStage>` with blocker description.
+
+### Priority 3: New session recommended
+
+If `context.recommendNewSession === true`:
+- Recommend: open a new Pi session.
+- Reason: The previous stage determined that context health + phase transition warrant a fresh session.
+- Output: copyable new-session prompt referencing `context.latestHandoffPath`, artifacts, and next stage.
+
+### Priority 4: Explicit next stage
+
+If `context.nextStage` exists, is meaningful, and differs from `context.currentStage`:
+- Recommend: `/skill:<context.nextStage>`.
+- Reason: The previous stage explicitly requested this transition.
+- Output: recommended skill command.
+
+### Priority 5: Stage mismatch detection
+
+If `context.currentStage` does not match the most recent artifact state (e.g. a plan exists but context says `"01-brainstorm"`):
+- Recommend: the skill that matches the most recent artifact, or ask the user to clarify.
+- Reason: Runtime context and artifact state are out of sync.
+- Output: suggested correction with explanation of the mismatch.
+
+### Priority 6: Fallback — artifact-count rules
+
+If none of the above context rules triggered, fall back to the artifact-count rules below.
 
 ## Rule 1: No brainstorm artifacts
 

package/skills/07-worktree/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 07-worktree
-description: Create and manage git worktrees for isolated feature development.
+description: "Create and manage git worktrees for isolated feature development. Use when starting a feature that needs branch isolation."
 ---
 
 # Worktree

package/skills/08-help/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: 08-help
-description: Explain when to use each Compound Engineering Phase 1 skill and how they connect.
+description: "Explain Phase 1 skills and their connections. Use when learning the workflow or deciding which skill applies."
 ---
 
 # Help

package/skills/references/pipeline-config.md

@@ -18,7 +18,34 @@ Supported `modelStrategy` formats in `.pi/settings.json`:
 - Full reference: `"02-plan": "anthropic/claude-opus-4-1"`
 - Bare model id (reuses current provider): `"02-plan": "claude-opus-4-1"`
 
-## End of skill: status + context
+## Start of skill: context loading
+
+Before reading any project files or running repository-wide scans, load the most recent handoff:
+
+1. Try `context_handoff load` or `context_handoff latest` first.
+2. Fallback: `read .context/compound-engineering/handoffs/latest.md` if tool is unavailable.
+3. **Handoff found?** Consume its `activeFiles`, `blocker`, `verification`, `currentTruth`, `activeRules` before any broad project reads.
+4. **No handoff?** Proceed normally — this is a new project or first run. Do not block.
+
+Core principle: **consume handoff before broad project file reads** — a single handoff read (~500 tokens) avoids 5-10 project file scans (~5K-10K tokens).
+
+## End of skill: save handoff + status + context
+
+Every Phase 1 skill (02-plan through 05-learn) must save context handoff at completion:
+
+```
+context_handoff save
+  currentStage: <stageKey>
+  nextStage: <next stage>
+  contextHealth: good | watch | heavy | critical
+  activeFiles: [1-5 currently active paths]
+  blocker: <blocker or N/A>
+  verification: <latest command + result>
+  activeRules: [1-5 rules critical for continuation]
+  currentTruth: [validated truths]
+```
+
+If `context_handoff` is unavailable, manually write the Handoff-lite template to `.context/compound-engineering/handoffs/latest.md`.
 
 Before final completion, always output these blocks (replace placeholders with real values, never output angle-bracket placeholders literally):