gm-cc 2.0.416 → 2.0.417

package/.claude-plugin/marketplace.json

@@ -4,7 +4,7 @@
     "name": "AnEntrypoint"
   },
   "description": "State machine agent with hooks, skills, and automated git enforcement",
-  "version": "2.0.416",
+  "version": "2.0.417",
   "metadata": {
     "description": "State machine agent with hooks, skills, and automated git enforcement"
   },

package/agents/gm.md

@@ -13,6 +13,7 @@ enforce: critical
 
 All work coordination, planning, execution, and verification happens through the skill tree:
 - `gm` skill → `planning` skill → `gm-execute` skill → `gm-emit` skill → `gm-complete` skill → `update-docs` skill
+- `memorize` skill — background only, non-sequential, launched via Agent with model=haiku and run_in_background=true
 
 All code execution uses `exec:<lang>` via the Bash tool — never direct `Bash(node ...)` or `Bash(npm ...)`.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gm-cc",
-  "version": "2.0.416",
+  "version": "2.0.417",
   "description": "State machine agent with hooks, skills, and automated git enforcement",
   "author": "AnEntrypoint",
   "license": "MIT",

package/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "gm",
-  "version": "2.0.416",
+  "version": "2.0.417",
   "description": "State machine agent with hooks, skills, and automated git enforcement",
   "author": {
     "name": "AnEntrypoint",

package/skills/gm/SKILL.md

@@ -121,6 +121,7 @@ Invoke `browser` skill. Escalation — exhaust each before advancing:
 **`gm-complete`** — VERIFY state. End-to-end verification and git enforcement. EXIT: invoke `gm-execute` if .prd items remain; invoke `update-docs` if .prd empty and pushed.
 **`update-docs`** — Refresh README, CLAUDE.md, and docs to reflect session changes. Invoked by `gm-complete`. Terminal state — declares COMPLETE.
 **`browser`** — Browser automation. Invoke inside EXECUTE state for all browser/UI work.
+**`memorize`** — Background memory agent (haiku, run_in_background=true). Launch when structural changes occur. Never blocks execution.
 
 ## PARALLEL SUBAGENTS (post-PLAN)
 
@@ -178,7 +179,7 @@ These rules apply to ALL states. Violations trigger immediate regression to PLAN
 - Adjacent step pairs are the most common failure site in chains — debug handoffs, not just individual steps
 
 **DOCUMENTATION** (update at every phase transition, not at the end):
-- CLAUDE.md: after each structural change, update technical info. NO progress/changelogs
+- CLAUDE.md: launch `memorize` subagent in background with what was learned. Never inline-edit CLAUDE.md directly.
 - TODO.md: add items when discovered, remove when done. File must not exist at completion
 - CHANGELOG.md: append entry after each commit
 - After push: deploy if deployable, publish if npm package

package/skills/gm-complete/SKILL.md

@@ -139,7 +139,7 @@ Before declaring complete, sweep the entire codebase for violations:
 8. **Fallback/demo modes** → remove, fail loud instead
 9. **TODO.md** → must be empty/deleted before completion
 10. **CHANGELOG.md** → must have entries for this session's changes
-11. **CLAUDE.md** → must reflect current technical state
+11. **memorize** → launch memorize subagent in background with session learnings before invoking update-docs
 12. **Deploy/publish** → if deployable, deploy. If npm package, publish.
 
 Any violation found = fix immediately before advancing.

package/skills/gm-emit/SKILL.md

@@ -86,7 +86,7 @@ The post-emit verification is a differential diagnosis against the pre-emit base
 - No adjectives/descriptive language in variable/function names
 - No unnecessary files — clean anything not required for the program to function
 - Client-side code exposes debug globals (window.__debug or similar)
-- CLAUDE.md reflects actual behavior
+- memorize subagent launched in background with what was learned before advancing
 - CHANGELOG.md updated with changes
 - TODO.md cleared or deleted
 

package/skills/gm-execute/SKILL.md

@@ -123,6 +123,14 @@ Real services, real data, real timing. Mocks/fakes/stubs/simulations = diagnosti
 
 **HYPOTHESIZE VIA EXECUTION — NEVER VIA ASSUMPTION**: Formulate a falsifiable hypothesis. Run it. Witness the output. The output either confirms or falsifies. Only a witnessed falsification justifies editing a file. Never edit based on unwitnessed assumptions — form hypothesis → run → witness → edit.
 
+## MEMORY
+
+When any mutable resolves from UNKNOWN to KNOWN (zero variance confirmed), launch memorize subagent in background — non-blocking, execution continues:
+
+`Agent(subagent_type='general-purpose', model='haiku', run_in_background=true, prompt=<memorize SKILL.md content>\n\n## CONTEXT TO MEMORIZE\n<resolved fact>)`
+
+Qualifies for memorization: new API shapes discovered, environment differences, behavioral constraints, runtime quirks, user feedback observed during execution.
+
 ## DO NOT STOP
 
 Never respond to the user from this phase. When all mutables are KNOWN, immediately invoke `gm-emit` skill. The chain continues until .prd is deleted and git is clean — that happens in `gm-complete`, not here.

package/skills/memorize/SKILL.md

@@ -0,0 +1,96 @@
+---
+name: memorize
+description: Background memory agent. Classifies and persists learned context into memory/ dir and CLAUDE.md. Run as haiku model in background only.
+allowed-tools: Read, Write, Bash(exec:nodejs)
+---
+
+# Memorize — Background Memory Agent
+
+You are invoked as a background Agent. You handle all memory persistence for this project.
+
+Memory dir: C:/Users/user/.claude/projects/C--dev-plugforge/memory/
+CLAUDE.md: C:/dev/plugforge/CLAUDE.md
+
+## STEP 1: READ CURRENT STATE
+
+Read memory dir contents and MEMORY.md index. Read all existing memory files. Read CLAUDE.md in full.
+
+If memory dir does not exist, create it. If MEMORY.md does not exist, create it empty.
+
+## STEP 2: CLASSIFY CONTEXT
+
+Examine the ## CONTEXT TO MEMORIZE section at the end of this prompt. Extract discrete facts. For each fact, classify as one of:
+
+- user: user role, goals, preferences, knowledge
+- feedback: guidance on approach — corrections AND confirmations
+- project: ongoing work, goals, bugs, incidents, decisions
+- reference: pointers to external systems, URLs, paths
+
+Discard:
+- Facts already covered in CLAUDE.md (any paraphrase or near-duplicate)
+- Obvious facts derivable from reading the code
+- Active task state, current progress, what was done in this session
+- Git history narration
+
+## STEP 3: CHECK CLAUDE.md FOR NON-OBVIOUS CAVEATS
+
+A non-obvious technical caveat is a fact that required multiple failed runs to discover — something that would not be apparent from reading the code or documentation.
+
+For each fact in context that qualifies:
+- If CLAUDE.md already covers it (exact or equivalent) → skip
+- If it is genuinely non-obvious and cost multiple runs → append to the appropriate section in CLAUDE.md
+
+Never add:
+- Obvious patterns visible in the code
+- Active task progress or session state
+- Redundant restatements of existing entries
+
+## STEP 4: WRITE MEMORY FILES
+
+For each classified fact:
+1. Check all existing memory files — if an existing file covers the same topic, merge the new fact into that file
+2. If no existing file covers it, create a new file: memory/<slug>.md
+
+File format:
+```
+---
+name: <descriptive name>
+description: <one-line description under 80 chars>
+type: user|feedback|project|reference
+---
+
+<body>
+```
+
+For feedback type, body must include:
+- The rule
+- Why: <reason>
+- How to apply: <concrete application>
+
+For project type, body must include:
+- The fact or decision
+- Why: <reason>
+- How to apply: <concrete application>
+
+## STEP 5: UPDATE MEMORY.md INDEX
+
+Rewrite the MEMORY.md index to reflect all current files in the memory dir.
+
+Format: one line per file, under 150 chars each:
+`- [Title](file.md) — one-line hook`
+
+No frontmatter. Max 200 lines.
+
+## STEP 6: CONSOLIDATE
+
+For each memory file: if its content is already fully covered by CLAUDE.md (exact or equivalent coverage), delete the memory file and remove its line from MEMORY.md.
+
+Do not delete memory files that cover topics CLAUDE.md does not address.
+
+## STEP 7: DONE
+
+No output required. Write files silently. Do not respond to the user.
+
+---
+
+## CONTEXT TO MEMORIZE

package/skills/update-docs/SKILL.md

@@ -50,7 +50,7 @@ Write only sections that changed. Do not rewrite unchanged content. Rules per fi
 
 **README.md**: platform count matches adapters in `platforms/`, skill tree diagram matches current state machine, quick start commands work.
 
-**CLAUDE.md**: Only non-obvious technical caveats that required multiple runs to discover — things that could not be known without hitting the problem first. Remove anything that no longer applies. Never add anything obvious from reading the code or that any developer would already know. The test: "would a developer need to discover this the hard way, or is it self-evident?" If self-evident, exclude it.
+**CLAUDE.md**: Launch memorize subagent in background with session learnings. Do not inline-edit CLAUDE.md — the memorize agent handles extraction, deduplication, and writing.
 
 **docs/index.html**: `PHASES` array matches current skill state machine phases. Platform lists match `platforms/` directory. State machine diagram updated if new phases added.
 
@@ -77,7 +77,7 @@ Witness each written file from disk. Any variance from expected content → fix
 
 ```
 exec:bash
-git add README.md CLAUDE.md docs/index.html plugforge-starter/agents/gm.md
+git add README.md docs/index.html plugforge-starter/agents/gm.md
 git diff --cached --stat
 ```