feat: port enrich skill, changes log, and brain-first lookup from gbrain

CHANGELOG.md

@@ -4,6 +4,18 @@ All notable changes to iContext will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [0.4.0] - 2026-05-04
+
+### Added
+- **`icontext-write-fact` skill.** Explicit decision tree that routes content to the correct vault location before writing. Covers six categories: legal/entity facts (`vault/legal/`), project facts (`vault/projects/`), team/people (`vault/team/`), strategy (`vault/strategy/`), secretarial activity (`vault/secretary/logs/`), and credentials. Includes a full top-level vault directory reference so agents can match against the real structure without guessing. Eliminates the recurring "agent dumps legal facts in log files" failure mode.
+- `icontext-populate-profile` now references `icontext-write-fact` for any durable facts surfaced during profile synthesis (e.g. company registrations, attorney contacts, funding dates).
+- `icontext-write-fact` ships with `icontext init` — installed to `~/.claude/skills/` and `~/.cursor/rules/` alongside the existing three skills.
+- CLAUDE.md snippet updated to list all four skills.
+- Tests for the new skill: frontmatter, decision tree category coverage, init installation, Cursor rule, skills-list count, and CLAUDE.md reference.
+
+### Changed
+- Bumped version to 0.4.0.
+
 ## [0.3.0] - 2026-05-02
 
 ### Added

cli.py

@@ -9,7 +9,7 @@
 import sys
 from pathlib import Path
 
-__version__ = "0.3.0"
+__version__ = "0.4.0"
 
 
 # ---------------------------------------------------------------------------
@@ -398,7 +398,12 @@ def _install_skills() -> tuple[int, list[str]]:
         msgs.append(_warn("skills/ source dir not found — skipping skill install"))
         return 0, msgs
 
-    skill_names = ["icontext-populate-profile", "icontext-refresh-profile", "icontext-share-card"]
+    skill_names = [
+        "icontext-populate-profile",
+        "icontext-refresh-profile",
+        "icontext-share-card",
+        "icontext-write-fact",
+    ]
     claude_skills_dir = Path("~/.claude/skills").expanduser()
     cursor_rules_dir = Path("~/.cursor/rules").expanduser()
     claude_skills_dir.mkdir(parents=True, exist_ok=True)
@@ -450,7 +455,8 @@ def _install_claude_md_snippet(vault: Path) -> None:
         "Available skills:\n"
         "- icontext-populate-profile  (build profile from Gmail/LinkedIn/chat)\n"
         "- icontext-refresh-profile   (update stale profile)\n"
-        "- icontext-share-card        (regenerate shareable summary)\n\n"
+        "- icontext-share-card        (regenerate shareable summary)\n"
+        "- icontext-write-fact        (route a fact to the correct vault location)\n\n"
         "Multi-device sync: at session start, run `icontext pull` to fetch any updates\n"
         "from other machines. The user-prompt-submit hook does this automatically if a\n"
         "remote is configured.\n"
@@ -518,7 +524,12 @@ def cmd_skills(args: argparse.Namespace) -> int:
     action = getattr(args, "skills_action", None) or "list"
     claude_skills_dir = Path("~/.claude/skills").expanduser()
     cursor_rules_dir = Path("~/.cursor/rules").expanduser()
-    skill_names = ["icontext-populate-profile", "icontext-refresh-profile", "icontext-share-card"]
+    skill_names = [
+        "icontext-populate-profile",
+        "icontext-refresh-profile",
+        "icontext-share-card",
+        "icontext-write-fact",
+    ]
 
     if action == "list":
         _header("skills")
@@ -1023,7 +1034,7 @@ def main() -> int:
             "  icontext connect linkedin --pdf ~/Downloads/Profile.pdf\n"
             "  icontext sync\n"
             "\n"
-            "docs: https://icontext.dev\n"
+            "docs: https://icontext.floom.dev\n"
             "issues: https://github.com/floomhq/icontext/issues"
         ),
     )

get.sh

@@ -37,7 +37,7 @@ if [ "${#MISSING_DEPS[@]}" -gt 0 ]; then
         echo "  sudo apt install git python3"
     fi
     echo ""
-    echo "Then re-run: curl -fsSL https://icontext.dev/install | bash"
+    echo "Then re-run: curl -fsSL https://icontext.floom.dev/install | bash"
     exit 1
 fi
 # -----------------------------------------------------------------------------

launch/COPY.md

@@ -25,7 +25,7 @@ Gemini-based headless sync is also available (`pip install icontext[sync]`) for
 
 Install:
 ```
-curl -fsSL icontext.dev/install | bash
+curl -fsSL icontext.floom.dev/install | bash
 icontext init
 ```
 Then: open Claude Code and say "Populate my icontext profile."
@@ -85,7 +85,7 @@ icontext is a folder and a set of skills.
 
 Two commands:
 ```bash
-curl -fsSL icontext.dev/install | bash
+curl -fsSL icontext.floom.dev/install | bash
 icontext init
 ```
 
@@ -186,7 +186,7 @@ The whole thing runs locally. No server, no cloud sync. The context folder is sp
 
 Works with Claude Code, Cursor, and Codex. Two install commands:
 
-curl -fsSL icontext.dev/install | bash
+curl -fsSL icontext.floom.dev/install | bash
 icontext init
 
 Open source, on GitHub: https://github.com/floomhq/icontext
@@ -218,7 +218,7 @@ Also works with Cursor (installs into `.cursor/rules/`) and Codex (via optional
 
 Install:
 ```
-curl -fsSL icontext.dev/install | bash
+curl -fsSL icontext.floom.dev/install | bash
 icontext init
 ```
 

launch/video-script-v2.txt

@@ -59,13 +59,13 @@ Three quick visual cards or screen recordings. Each one beat.
 
 
 [1:05 — 1:15]  CTA
-Visual: Install command big and clean. icontext.dev URL.
+Visual: Install command big and clean. icontext.floom.dev URL.
 
   "icontext. Open source.
 
   Two commands. No API key. Your AI never forgets you again.
 
-  icontext.dev"
+  icontext.floom.dev"
 
 (end on the URL on screen, hold 2 seconds)
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.backends.legacy:build"
 
 [project]
 name = "icontext"
-version = "0.3.0"
+version = "0.4.0"
 description = "Persistent memory for AI coding tools — local, encrypted, built from your real data."
 readme = "README.md"
 license = {text = "MIT"}

skills/fbrain-enrich/SKILL.md

@@ -0,0 +1,58 @@
+---
+name: fbrain-enrich
+description: >
+  Update a single person or project entry in the fbrain vault from the current
+  conversation. Use when the user says "I just spoke to X", "update X in my
+  brain", "enrich X", or "add note about X".
+---
+
+# fbrain: Enrich One Entry
+
+Update one existing person or project entry without rerunning populate or refresh.
+
+## Trigger
+
+Use this skill when the user asks to enrich, update, or add a note about one
+person or project.
+
+## Target
+
+1. Identify the target name from the user's message.
+2. Read `~/context/internal/profile/relationships.md` and
+   `~/context/internal/profile/projects.md`.
+3. Match the target against exactly one existing person or project entry.
+4. If there is no clear match, ask one short clarification question.
+
+## Source
+
+Use only the current conversation context:
+
+- the user's message
+- recent tool outputs in this session
+- facts the user just confirmed
+
+Do not call external APIs, search the web, read Gmail, browse LinkedIn, or run
+the full populate or refresh pipeline.
+
+## Write
+
+Update only the matched entry in place:
+
+- For a person, edit just that row in `relationships.md`.
+- For a project, edit just that project line or paragraph in `projects.md`.
+- Preserve table formatting, headings, ordering, and every unrelated entry.
+- Keep the change short and evidence-grounded.
+
+## Log
+
+Append one line to `~/context/internal/changes.md`:
+
+```text
+YYYY-MM-DD: enriched <name> — <one-line summary of what changed>
+```
+
+Create the file if it does not exist.
+
+## Output
+
+Confirm what changed in one sentence.

skills/icontext-populate-profile/SKILL.md

@@ -145,6 +145,13 @@ generated: YYYY-MM-DD
 
 Strict 200-word max. No corporate filler. No "passionate about", no "pioneering". Plain second person. The user's actual voice from their actual data.
 
+## Changelog
+
+After writing all four profile files, append one line to `~/context/internal/changes.md` (create the file if it does not exist):
+```
+YYYY-MM-DD: populate — source: <source used>, <N> relationships, <M> projects
+```
+
 ## Voice rules
 
 - Plain second person ("you are", "you work on")
@@ -163,6 +170,14 @@ If `internal/profile/user.md` exists:
 - If 7-30 days old: ask the user "your profile is N days old, refresh?"
 - If >30 days old: refresh automatically.
 
+## Saving facts that surface during synthesis
+
+If during synthesis you encounter a durable fact that doesn't belong in the
+profile (e.g. the user mentions their company's Delaware registration, an
+attorney contact, a funding round closing date), use the `icontext-write-fact`
+skill to route it to the correct vault location. Do not dump it into the
+profile files or a log file.
+
 ## Output
 
 After writing, summarize for the user:
@@ -171,3 +186,12 @@ After writing, summarize for the user:
 - Number of relationships, projects, topics extracted
 - File paths written
 - Suggested next: "Open a new Claude Code session and ask me what I know about you."
+
+## Injecting into CLAUDE.md
+
+After running this skill, paste the following block into your CLAUDE.md to enable brain-first lookups:
+
+```
+## fbrain: brain-first lookup
+Before searching the web or calling external APIs for information about a person, company, or project, check ~/context/internal/profile/relationships.md and ~/context/internal/profile/projects.md first. If a relevant entry exists, use it as the primary source and supplement from external sources only if the entry is stale or incomplete.
+```

skills/icontext-refresh-profile/SKILL.md

@@ -70,6 +70,13 @@ previous_generated: {prior_date_from_old_file}
 
 This gives future refresh runs a trail of when the profile last meaningfully changed.
 
+## Changelog
+
+After a write (skip if no meaningful changes), append one line to `~/context/internal/changes.md`:
+```
+YYYY-MM-DD: refresh — <changed/no change>, source: <source>
+```
+
 ## Output
 
 Summarize for the user: