feat: add research vault recall to Algorithm, Research, and WorldThreatModel workflows

Releases/v4.0.3/.claude/PAI/Algorithm/v3.7.0.md

@@ -147,6 +147,8 @@ The effort level defaults to `standard` here and gets refined later in OBSERVE a
 
 **FIRST ACTION:** Voice announce `"Entering the Observe phase."`, then Edit PRD frontmatter `updated: {timestamp}`. Then thinking-only, no tool calls except context recovery (Grep/Glob/Read <=34s)
 
+- **RESEARCH VAULT CHECK (all tasks):** Scan `MEMORY/RESEARCH/` for prior research relevant to the current request. List files (`ls MEMORY/RESEARCH/`), read file names and frontmatter titles, match against request keywords. If relevant research exists, note it in the ISC context and include it as a resource for subsequent phases. If no relevant research exists, note "no relevant prior research in vault" and move on. This is a fast check — `ls` + at most 2-3 frontmatter reads, no spawned agents.
+
 - REQUEST REVERSE ENGINEERING: explicit wants, implied wants, explicit not-wanted, implied not-wanted, common gotchas, previous work
 
 OUTPUT:

Releases/v4.0.3/.claude/skills/Research/Workflows/DeepInvestigation.md

@@ -51,6 +51,12 @@ Read `~/.claude/MEMORY/STATE/current-work.json` for the active work directory.
 ### Step 0: Detect Iteration State
 
 ```
+CHECK broader research vault for related prior work:
+  - List all directories/files in MEMORY/RESEARCH/
+  - If OTHER research topics (not this investigation's vault) have keyword overlap,
+    note them as supplementary context for landscape and entity research
+  - Include relevant prior findings in agent briefs during Steps 1 and 4
+
 READ vault directory for existing artifacts:
   - LANDSCAPE.md exists? → This is a CONTINUATION (skip to Step 3 or 4)
   - ENTITIES.md exists? → This is a CONTINUATION (skip to Step 3 or 4)

Releases/v4.0.3/.claude/skills/Research/Workflows/ExtensiveResearch.md

@@ -21,9 +21,26 @@ See `SKILL.md` for full URL Verification Protocol.
 
 ## Workflow
 
-### Step 0: Generate Creative Research Angles (deep thinking)
+### Step 0: Check Research Vault for Prior Work
 
-**Use deep thinking to generate diverse research angles:**
+Before generating angles, check if the research vault already has relevant coverage:
+
+```
+1. List files in MEMORY/RESEARCH/ (recursive, .md files only)
+2. Match file names and directory names against the research topic keywords
+3. If matches found: read the matched file's frontmatter/first 20 lines for a summary
+```
+
+**If relevant prior research exists:**
+- Use prior findings to inform angle generation in Step 0.5 — avoid re-researching known territory, focus angles on gaps
+- Include a brief summary in each agent's prompt: "Prior research on this topic found: [summary]. Focus on NEW findings beyond what's already known."
+- If prior research is comprehensive and recent (<30 days), suggest: "We have existing research on this topic from [date]. Want to extend it rather than start fresh?"
+
+**If no relevant research:** Proceed normally.
+
+### Step 0.5: Generate Creative Research Angles (deep thinking)
+
+**Use deep thinking to generate diverse research angles** (informed by any prior vault findings from Step 0):
 
 Think deeply about the research topic:
 - Explore multiple unusual perspectives and domains

Releases/v4.0.3/.claude/skills/Research/Workflows/QuickResearch.md

@@ -11,6 +11,22 @@
 
 ## Workflow
 
+### Step 0.5: Check Research Vault for Prior Work
+
+Before spawning the agent, do a fast check of the research vault:
+
+```
+1. List files in MEMORY/RESEARCH/ (recursive, .md files only)
+2. Match file names and directory names against the research topic keywords
+3. If matches found: read the matched file's frontmatter/first 20 lines for a summary
+```
+
+**If relevant prior research exists:**
+- Include a brief summary in the agent's prompt: "Prior research on this topic found: [summary]. Build on these findings."
+- If prior research is comprehensive and recent (<30 days), suggest: "We have existing research on this topic from [date]. Want to extend it rather than start fresh?"
+
+**If no relevant research:** Proceed normally.
+
 ### Step 1: Launch Single Claude Agent
 
 **ONE Task call - Claude researcher with a single focused query:**

Releases/v4.0.3/.claude/skills/Research/Workflows/StandardResearch.md

@@ -20,6 +20,22 @@ See `SKILL.md` for full URL Verification Protocol.
 
 ## Workflow
 
+### Step 0.5: Check Research Vault for Prior Work
+
+Before spawning agents, check if the research vault already has relevant coverage:
+
+```
+1. List files in MEMORY/RESEARCH/ (recursive, .md files only)
+2. Match file names and directory names against the research topic keywords
+3. If matches found: read the matched file's frontmatter/first 20 lines for a summary
+```
+
+**If relevant prior research exists:**
+- Include a brief summary in each agent's prompt as context: "Prior research on this topic found: [summary]. Build on these findings."
+- If prior research is comprehensive and recent (<30 days), suggest to the user: "We have existing research on this topic from [date]. Want to extend it rather than start fresh?"
+
+**If no relevant research:** Proceed normally.
+
 ### Step 1: Craft One Query Per Researcher
 
 Create ONE focused query optimized for each researcher's strengths:

Releases/v4.0.3/.claude/skills/Thinking/WorldThreatModelHarness/Workflows/TestIdea.md

@@ -44,6 +44,23 @@ curl -s -X POST http://localhost:8888/notify \
   -d '{"message": "Testing your idea against all eleven world threat models at TIER tier", "voice_id": "fTtv3eikoepIosk8dTZ5"}'
 ```
 
+### Step 1.5: Check Research Vault for Relevant Prior Work
+
+Before decomposing the idea, check if the research vault has relevant prior research:
+
+```
+1. List files in MEMORY/RESEARCH/ (recursive, .md files only)
+2. Match file names and directory names against the idea's domain keywords
+3. If matches found: read the matched file's frontmatter/first 20 lines for a summary
+```
+
+**If relevant prior research exists:**
+- Summarize the key findings from prior research
+- Include this summary as grounding context in Step 3's agent briefs: "Prior research on this domain found: [summary]. Use these findings as factual grounding when testing the idea against your time horizon."
+- This gives world model agents concrete evidence instead of relying solely on training knowledge
+
+**If no relevant research:** Proceed normally — note "no prior research vault data for this domain."
+
 ### Step 2: Extract and Decompose the Idea
 
 Before hitting it with world models, decompose the idea: