create-egregore 0.19.8 → 0.19.11

package/runtime/codex/.codex/skills/deep-reflect/SKILL.md

@@ -1,74 +1,109 @@
 ---
 name: deep-reflect
-description: Cross-reference an insight against the Egregore knowledge base using Codex-native staged analysis when the user invokes /deep-reflect or $deep-reflect.
+description: Deep research over the Egregore org memory — multi-hop, evidence-verified synthesis of what the org collectively knows — when the user invokes /deep-reflect or $deep-reflect.
 ---
 
-# Egregore Deep Reflect
+# Egregore Deep Reflect (v4)
 
-Native Codex Egregore skill. Use this for evidence-based analysis of how a new
-insight relates to the existing knowledge base.
+Native Codex Egregore skill. Ask the org's memory a question and return a
+verified, cited synthesis of what it collectively knows — including what it
+doesn't. `/reflect` captures user thought; `/deep-reflect` researches what the
+org already knows.
 
-## Modes
+Full contracts live in `.claude/skills/deep-reflect/` (`SKILL.md` entry,
+`RESEARCH.md` engine, `ROLES.md` agent roles, `REPORT.md` output). Follow them;
+this file is the Codex-runtime adapter.
 
-- empty request: deep mode.
-- `focused ...`: focused topic mode.
-- `quick ...` or `category: content`: quick mode.
+## Modes
 
-Detect directed questions when the user asks how, why, what something means, or
-about a relationship. Preserve the user's lens while still surfacing important
-signals outside the lens.
+- Question (default): interrogative or bare-topic arguments → "what does the
+  org know about X?"
+- Cross-ref: "cross-reference this" / a declarative insight → "how does this
+  insight sit against existing memory?"
+- Depth: `--brief` (1 hop wave), standard (≤3), `--deep` (≤5). Default is
+  dynamic from the corpus probe.
 
 ## Flow
 
-1. Check `egregore.json` mode first. In local mode, do a lightweight reflection
-   from memory files and do not call graph, batch graph, or notification
-   scripts.
-2. In all modes, run `bin/agent.sh search "$TOPIC" -n 6` — hybrid content
-   recall over the whole memory repo. Feed strong hits into the candidate
-   set; in local mode this is the primary context source.
-3. In connected mode, additionally gather graph context with `bin/graph.sh`
-   and suppress raw JSON. Query recent sessions, active quests, recent
-   artifacts, decisions, the topic landscape, and the full artifact
-   landscape.
-4. If the graph is unavailable or has fewer than about ten artifacts, fall back
-   to lightweight reflection and say why in one line.
-5. Select candidate artifacts. Prefer title, type, topics, quest, author, and
-   date metadata first; fetch file contents only for the selected subset.
-6. Use Codex multi-agent or subagent tooling only if it is available in the
-   current session. If not, run a single-model staged analysis:
-   - candidate selection summary
-   - evidence extraction summary
-   - signal analysis
-   - final synthesis
-7. Classify signals as tension, convergence, gap, dependency, phase shift,
-   redundancy, emergence, reinforcement, or a named custom type.
-8. Write a markdown artifact under `memory/knowledge/deep-reflect/` with
-   frontmatter for run id, mode, lens, author, created time, and selected
-   evidence paths.
-9. In connected mode, write best-effort graph edges with `bin/graph.sh` or
-   `bin/graph-batch.sh`. Use `TENSION_WITH` for conflict signals and
-   `RELATES_TO` for other signals.
-10. Save memory and repo changes with `bin/agent.sh save --message "Deep reflect: $TOPIC" --topic "$TOPIC"`.
+1. Check `egregore.json` mode FIRST. Local mode: full engine over files; never
+   call graph, batch-graph, or notification scripts; zero graph vocabulary in
+   output. Connected: graph adds hop rails and writeback, every graph call
+   non-fatal (`2>/dev/null || true`).
+2. FRAME: corpus probe — 2× `bin/agent.sh search "$TOPIC" --fast -n 20` with
+   different wordings + `bin/artifacts.sh find "$TOPIC"` +
+   `grep -ril "$TOPIC" memory/knowledge/ memory/handoffs/ memory/artifacts/`.
+   N = the union of unique paths; set the wave budget from N per
+   `RESEARCH.md` §Budget; decompose the question into 2-4 sub-questions;
+   print a plan line (`Researching: … · ~N relevant docs · budget M waves ·
+   est ~X min`). Under ~5 relevant docs: say memory is too thin and offer
+   `/reflect` or `/harvest` instead of faking research. If search errors or
+   returns nothing while grep hits, rely on grep and note "recall degraded".
+3. SEED: check `memory/knowledge/research/` for prior runs on the topic —
+   their cited artifacts (frontmatter `artifacts_cited`, older reports use
+   `artifacts_consulted`; resolve filename stems with `find`) are pre-warmed
+   leads, and their `null_regions` gaps get re-audited. Build a frontier of
+   ≤12 docs from search probes, frontmatter topic/quest greps, and index
+   ledgers.
+4. HOP WAVES: the Codex runtime has no Workflow tool — run the foreground
+   staged path from `ROLES.md`: per wave, read 8-12 docs (subagents if the
+   session offers them, otherwise staged single-model passes), extract
+   claims-with-verbatim-excerpts and typed leads (doc links, quests, people,
+   sessions, topic terms, named absences). Dedup against the seen-set, score
+   leads, hop until saturation per `RESEARCH.md` (coverage on every
+   sub-question plus low novelty or zero yield; or budget).
+5. PATTERN: two de-correlated passes. Without subagents true blindness is
+   impossible — degrade deliberately: run the TENSION pass FIRST (grouped by
+   topic), then the convergence pass (chronological), so tension-finding is
+   never anchored by an existing convergence story. Disagreement between
+   passes is a finding, never averaged away.
+6. VERIFY: for every claim, whitespace-normalize the excerpt and re-find it in
+   the cited file (miss = miscite = kill; supported paraphrase = correct the
+   excerpt). The mechanical re-check survives self-verification — it is a
+   grep, not a judgment. Re-attack every gap claim with one extra probe; a
+   found doc refutes the gap and joins the ledger. Evidence older than ~90
+   days gets one `--fast` newer-artifact probe. Report the kill count. When
+   the verifying context is the same one that made the claims, mark the
+   report's provenance `verification: self-checked`.
+7. SYNTHESIZE + PRESENT: stance, not summary — answer first, then ◆ primary /
+   ◇ secondary findings with `memory/...` path citations, tensions, and "what
+   memory doesn't hold" (gaps + the probes that prove them). Confidence is
+   derived from evidence structure (verified + ≥3 independent sources = high;
+   single-source always labeled). Register: analyst, not critic.
+8. CAPTURE (user-gated): Save (default) / Edit first / File follow-ups /
+   Skip. On Save write `memory/knowledge/research/YYYY-MM-DD-author-slug.md`
+   with frontmatter (question, mode, stop_reason, waves, docs_read,
+   claims_killed, builds_on, artifacts_consulted, artifacts_cited,
+   null_regions), then connected-mode best-effort graph writes per the main
+   `REPORT.md` §Writeback (MERGE with filePath — do NOT use
+   `graph-op.sh register-artifact`; `RELATES_TO`/`TENSION_WITH` for verified
+   cross-ref claims only), then
+   `bin/agent.sh save --message "Deep research: $TOPIC" --topic "$TOPIC"`.
+   On Skip: nothing is written except the local run ledger under
+   `.egregore/research-runs/`. Render the TUI box AFTER the gate, reflecting
+   the actual outcome.
 
 ## Output
 
-Lead with the strongest finding, then list primary signals, evidence paths, and
-recommended next action. Keep secondary and ambient signals short.
-
-Structured UX parity is required. Then render the Egregore deep-reflect
-confirmation TUI:
+Render the Egregore deep-reflect confirmation TUI after the capture gate:
 
-- Use a 72-column outer box with standard top/separator/content/bottom lines.
-- Header: `DEEP REFLECT`, author, and date.
-- Body: lens/topic, strongest finding, signal counts by type, selected evidence
-  count, artifact path, and recommended next action.
-- Footer: saved/pushed state. In connected mode include graph-link state when
-  available; in local mode omit graph language.
-- Do not show raw graph JSON or replace the confirmation box with plain prose.
+- 72-column outer box, standard top/separator/content/bottom lines only.
+- Header: `◈ DEEP REFLECT`, author, date.
+- Body: question, ◆/◇ finding counts, confidence, waves · docs read · cited ·
+  claims cut in verification.
+- Footer: actual saved/pushed state; say "graphed" only if the graph write
+  succeeded; in local mode omit graph language entirely.
+- Structured UX parity is required: preserve the rendered deep-reflect TUI
+  box, no preamble, no prose-only replacement, and no raw search or graph
+  output.
+- Never show raw graph JSON; never replace the box with prose.
 
 ## Rules
 
-- Never show raw graph JSON.
-- Do not claim evidence you did not inspect.
-- Keep intermediate summaries explicit so the user can audit the reasoning.
+- Search budget: at most ONE non-`--fast` search probe per run (seed only);
+  `--fast` everywhere else. Count result blocks, never the banner hit count.
+- No claim without a verbatim excerpt that survives re-finding in the file.
+- Absence is a first-class finding — cite the probes that prove it.
+- Don't manufacture significance: "memory holds nothing structural on this"
+  is a valid report.
+- At most ONE clarifying question before the run, plus the capture gate.
 - Do not use Claude Code commands.

package/runtime/codex/.codex/skills/deploy-bridge/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: deploy-bridge
+description: 'Deploy the Telegram bridge (egregore_bridge.py) to the VPS agent box and verify it comes back healthy. Org-internal — the bridge does NOT auto-deploy from develop.'
+---
+
+<!-- generated-by: bin/codex-sync-skills.sh -->
+
+# Egregore deploy-bridge Adapter
+
+This is fallback coverage for a long-tail Egregore workflow that has not been
+ported to a hand-written Codex-native skill yet.
+
+Use the project shell and filesystem directly. Do not invoke Claude Code
+commands. Translate interactive choices to structured Codex question tooling
+when it is available; otherwise render compact numbered choices with an
+`Other:` option and wait for the user.
+
+1. Read `.claude/skills/deploy-bridge/SKILL.md` for the workflow details.
+2. Run the referenced `bin/` scripts directly from Codex.
+3. Treat graph, publish, and notify steps as best-effort unless that workflow
+   explicitly says they are required.
+4. Keep local-mode behavior filesystem-first and avoid graph or notification
+   calls when `egregore.json` declares `"mode": "local"`.
+5. Never call the deprecated `egregore-handoff` CLI for Egregore project
+   handoffs.

package/runtime/codex/.codex/skills/handoff/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: handoff
-description: Create an internal Egregore team handoff when the user invokes /handoff or $handoff, says they are done, asks to hand off work, or wants to pass context to a teammate or future self.
+description: Create an internal Egregore team handoff when the user invokes /handoff or $handoff, asks to hand off work, or wants to pass context to a teammate or future self.
 ---
 
 # Egregore Handoff
@@ -11,6 +11,10 @@ the Egregore scripts. External runnable packets are emissaries; if the user
 asks for one or provides an `egregore.xyz/emissary/e/` link, use the emissary
 tooling instead.
 
+`/handoff` is pure addressed capture. Do not triage incoming handoffs here;
+route “show/close my handoffs” to `$activity`. Personal session closure without
+an addressed recipient belongs to `$wrap`.
+
 ## Flow
 
 1. Run `bin/agent.sh sync`.
@@ -29,7 +33,7 @@ tooling instead.
 
 ```bash
 TMPDIR="$(mktemp -d -t codex-handoff-XXXXXX)"
-TMPDIR="$TMPDIR" bash bin/handoff-run.sh --author "$AUTHOR" --topic "$TOPIC" ${RECIPIENT_ARG} < "$BODY_FILE"
+TMPDIR="$TMPDIR" bash bin/capture-run.sh --mode addressed --author "$AUTHOR" --topic "$TOPIC" ${RECIPIENT_ARG} < "$BODY_FILE"
 ```
 
 Use `--recipient "$RECIPIENT"` only when a clear teammate or future-self

package/runtime/codex/.codex/skills/save/SKILL.md

@@ -17,11 +17,29 @@ which branch, commit, push, pull request, or memory sync steps are required.
 ```bash
 git status --short
 git status --short memory 2>/dev/null || true
-git branch --show-current
+BRANCH=$(git branch --show-current)
+AHEAD=0
+OPEN_PR=""
+case "$BRANCH" in
+  dev/*|feature/*|bugfix/*)
+    git fetch origin develop --quiet 2>/dev/null || true
+    AHEAD=$(git rev-list --count "origin/develop..HEAD" 2>/dev/null || echo 0)
+    if command -v gh >/dev/null 2>&1; then
+      OPEN_PR=$(gh pr list --head "$BRANCH" --base develop --state open \
+        --json url --jq '.[0].url // empty' 2>/dev/null || true)
+    fi
+    ;;
+esac
 ```
 
-3. If there are no repo or memory changes, say everything is already saved and
-   stop.
+3. A clean working tree alone does **not** mean the work is fully saved:
+   - If there are repo or memory changes, continue.
+   - If the task branch is ahead of `origin/develop` and `OPEN_PR` is empty,
+     continue even when the tree is clean. The bridge must push any committed
+     work and create the missing PR.
+   - Stop with "everything is already saved" only when there are no repo or
+     memory changes and either the branch has no commits to integrate
+     (`AHEAD=0`) or an open integration PR already exists.
 4. Synthesize:
    - a short topic from the work,
    - a clear commit message,

package/runtime/codex/.codex/skills/scroll/SKILL.md

@@ -18,7 +18,7 @@ small fork. A scroll requires all three legs: face, harvest, and update loop.
 ## Structured UX parity
 
 The browser artifact is the primary output. Preserve both tabs, deep links from
-honest costs into decisions, the versioned copydock fence, stable URL, and local
+honest costs into decisions, the versioned copydock return block, stable URL, and local
 source path. Do not substitute a prose-only summary or expose raw publish output.
 
 ## Route the invocation
@@ -121,13 +121,64 @@ the same `--id`; never mint a new URL for an update. In local mode, publishing
 is unavailable: open the file locally and report the source path without
 suggesting configuration changes.
 
-Add the URL or local path, source location, version, and fence tag to the
+Add the URL or local path, source location, version, and return-block tag to the
 relevant quest's log/artifacts tables. If no quest is relevant, keep the source
 path as the durable record and state it.
 
+## Creator gate — before any absorb
+
+The scroll record `memory/scrolls/<slug>.md` names `creator: <person-slug>` and
+optionally `trusted: [<person-slug>, …]`. A record missing `creator:` is
+backfilled from unambiguous existing attribution, otherwise ask the user once
+and write it — never silently default the gate open.
+
+Derive the paster from the session person (`.egregore-state.json`, then git
+config) — never from the return block's self-declared `respondent:` line, which
+is attribution metadata, not an authorization credential.
+
+If the paster is not the creator and not in `trusted:`, the return is **gated**:
+
+1. Append a `turn-received` event to `memory/scrolls/.events/<slug>.jsonl`
+   (create the file if absent). Event ids are append-if-absent and MUST match
+   the Claude corpus exactly — both runtimes share one event log:
+   `turn.id = respondent + '|' + base_version + '|' + date + '|' + digest`;
+   digest = FNV-1a 32-bit hex over the UTF-8 bytes of
+   `JSON.stringify([respondent, String(base_version), date, sortedAnswers,
+   sortedOpens])` where `sortedAnswers` = answers sorted by fork id, each
+   mapped to the tuple `[fork, pick-or-UNDECIDED, note-or-empty]` (labels are
+   excluded), and `sortedOpens` = opens sorted by prompt id, each mapped to
+   `[prompt, text-or-UNDECIDED]`; every string is NFC-normalized and
+   CRLF→LF-normalized before hashing. Record `submitted_by: <paster>` on the
+   event (event-log metadata only — never projected into a published ledger)
+   and `merged_into: "pending"`.
+2. Do NOT absorb: no `memory/knowledge/decisions/` writes, no spec/plan edits,
+   no face rewrite, no version bump, no republish of new content. Republishing
+   the unchanged current version to acknowledge the turn is allowed and appends
+   no version event.
+3. Notify the creator — connected: `bash bin/notify.sh send <creator>
+   "<paster> answered your scroll <slug> (vN): <digest> — pending your review"`;
+   local: write an addressed handoff in `memory/handoffs/` naming the scroll,
+   paster, base version, digest, and the dispositions below. The contributor
+   may also send the artifact link directly.
+4. Report the gated state and stop.
+
+**Creator review.** The creator disposes in chat: **accept** (absorb into the
+next version), **decline — <reason>** (mandatory nonblank note; refuse a
+decline without one), per-answer overrides (top-level disposition is the
+default, listed answers override; a turn with zero effectively accepted answers
+is a plain decline; a zero-answer turn takes the top-level disposition), or
+**synthesize** (compose one proposal from all pending turns — consensus
+decides, conflicts become new explicit forks — and publish only on creator
+approval). Every disposition appends `turn-reviewed`
+`{type, turn, disposition: accepted|declined, by, date, note?, answers?}` with
+id `review|<turn.id>` before any composition. An accepted turn stays pending
+until a version actually absorbs it. Declined turns are never absorbed, stay
+attributed in the log, and their reason travels with them. The creator's own
+returns and trusted respondents keep the immediate loop below.
+
 ## Update loop
 
-On a returned fence:
+On a returned block (creator or trusted paster — gated returns stop above):
 
 1. **Absorb.** Parse Q/O lines and notes. Write durable decisions with their
    reasoning to `memory/knowledge/decisions/`, update the underlying spec or

package/runtime/codex/.codex/skills/task/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: task
+description: 'Create, import, inspect, approve, answer, or cancel work in the hosted Egregore Agent Desk. Use for `/task`, `$task`, “send this work to the agents,” “start these tasks,” Markdown development batches, or checking hosted task progress. This skill submits work to the hosted pipeline; it does not implement the task in the current local session.'
+---
+
+<!-- generated-by: bin/codex-sync-skills.sh -->
+
+# Egregore task Adapter
+
+This is fallback coverage for a long-tail Egregore workflow that has not been
+ported to a hand-written Codex-native skill yet.
+
+Use the project shell and filesystem directly. Do not invoke Claude Code
+commands. Translate interactive choices to structured Codex question tooling
+when it is available; otherwise render compact numbered choices with an
+`Other:` option and wait for the user.
+
+1. Read `.claude/skills/task/SKILL.md` for the workflow details.
+2. Run the referenced `bin/` scripts directly from Codex.
+3. Treat graph, publish, and notify steps as best-effort unless that workflow
+   explicitly says they are required.
+4. Keep local-mode behavior filesystem-first and avoid graph or notification
+   calls when `egregore.json` declares `"mode": "local"`.
+5. Never call the deprecated `egregore-handoff` CLI for Egregore project
+   handoffs.

package/runtime/codex/.codex/skills/tui-design/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: tui-design
-description: 'Fallback Codex adapter for the Egregore tui-design workflow.'
+description: 'TUI Design System for Egregore slash commands. Use when designing or modifying any command''s terminal output to ensure visual consistency. Triggers on: TUI, terminal output, box drawing, command output formatting, sigils, section markers. Do NOT use for browser rendering (use dark-mode) or external prose (use egregore-voice).'
 ---
 
 <!-- generated-by: bin/codex-sync-skills.sh -->

package/runtime/codex/.codex/skills/wrap/SKILL.md

@@ -6,7 +6,10 @@ description: Wrap an Egregore session when the user invokes /wrap or $wrap, or s
 # Egregore Wrap
 
 Native Codex Egregore skill. Use this inside an Egregore checkout with
-`bin/agent.sh`.
+`bin/agent.sh`. The command routes through `bin/capture-run.sh --mode personal`.
+It writes the personal record first, appends graph and handoff-completion events
+to the local WAL, and reconciles them detached; do not wait for or repeat that
+background work.
 
 ## Flow
 

package/runtime/codex/.codex/spec-manifest.json

@@ -14,6 +14,18 @@
       "sourceHash": "ddcbc0083aff",
       "note": "Kept; /update, /contribute, /save tokens rendered as $skill tokens."
     },
+    {
+      "section": "Cross-Runtime Compatibility",
+      "action": "keep",
+      "sourceHash": "75d02e2603f5",
+      "note": "Kept verbatim — framework parity applies equally to Claude Code, Codex, and Pi."
+    },
+    {
+      "section": "Voice",
+      "action": "keep",
+      "sourceHash": "af73d4eaa7cd",
+      "note": "Kept — all runtimes can consult the shared voice rules and register-specific skills."
+    },
     {
       "section": "On Launch — MANDATORY FIRST ACTION",
       "action": "replace",
@@ -23,7 +35,7 @@
     {
       "section": "After Greeting — BRANCH ON FIRST RESPONSE",
       "action": "replace",
-      "sourceHash": "c71340e68c9e",
+      "sourceHash": "bce266fb848b",
       "note": "EnterWorktree -> bin/agent.sh branch with git checkout fallback; AskUserQuestion -> numbered options in plain text; branch-guard mapped to .codex/hooks/branch-guard.js with instruction fallback; plan-mode note dropped (no plan mode on Codex); /branch + /onboarding -> skill tokens. Handoff claiming and repoState auto-checkout kept verbatim (pure git)."
     },
     {
@@ -35,7 +47,7 @@
     {
       "section": "Knowledge Graph",
       "action": "keep",
-      "sourceHash": "5c0953aee955",
+      "sourceHash": "b5d9a149f1a6",
       "note": "Kept verbatim — bin/graph.sh is runtime-neutral."
     },
     {
@@ -83,7 +95,7 @@
     {
       "section": "Command Awareness",
       "action": "keep",
-      "sourceHash": "e1db878ccaee",
+      "sourceHash": "ce5eb3015427",
       "note": "Prepended the Codex command surface (skill tokens, native vs adapter split, $save abstraction); disambiguation map kept with /x -> $x token rendering. Tokens with no Codex skill (e.g. /loop, a Claude Code harness feature) are left as-is."
     },
     {

package/runtime/codex/AGENTS.md

@@ -96,6 +96,24 @@ When a user reports broken or missing framework behavior, the first question is
 
 ---
 
+## Cross-Runtime Compatibility
+
+Framework behavior must work across Claude Code, Codex, and Pi Egregores. Put
+shared mechanics in runtime-neutral `bin/` scripts; keep `.claude/`, `.codex/`,
+and `.pi/` as thin adapters to those mechanics. When changing startup, session
+lifecycle, capture, skills, or commands, trace all three runtime entry points
+and add or update a parity test. Do not call a framework change complete based
+on one runtime alone. Regenerate derived Codex and Pi specifications or runtime
+bundles whenever their sources change.
+
+---
+
+## Voice
+
+Egregore's voice is governed by `.claude/rules/voice-bedrock.md` (always loaded). Register-specific skills: `egregore-voice` (external), `product-voice` (internal UX), `character-v4` (encounters), `alpha-openers` (outreach). See voice-bedrock for the full register map.
+
+---
+
 ## On Launch — MANDATORY FIRST ACTION
 
 The `egregore` launcher renders the Egregore startup card (identity, team momentum, and pending work) via `bin/codex-session-start.sh` before Codex starts, and installs project skills. Do not rerun startup checks and do not narrate startup. The card ends with **"What are you working on?"** — that question is already on screen; treat the user's first message as the answer to it. To re-show the card, run `bash bin/codex-session-start.sh --card`.
@@ -146,19 +164,22 @@ If on develop after two messages, create a branch immediately from whatever cont
 
 ### Branch-guard protocol
 
-The `.codex/hooks/branch-guard.js` PreToolUse hook (enabled by the launcher via `--enable hooks`) denies writes, patches, and commits on `develop`/`main`/`master`. When it fires, choose between two paths based on how clear the topic is:
+The `.codex/hooks/branch-guard.js` PreToolUse hook (enabled by the launcher via `--enable hooks`) protects project writes on `develop`/`main`/`master`. Its block message is operational guidance, not a reason to interrupt the user with routine Git choices:
 
-- **Topic is obvious from the last turn** (user said "fix X", "add Y feature") → auto-branch silently with `bin/agent.sh branch --topic "<topic>"` (or the git checkout fallback), then continue. Say one sentence ("Creating `dev/{author}/{slug}`.") so the user isn't confused about why a new branch appeared.
-- **Topic is ambiguous** (open-ended exploration, user didn't specify what we're building yet) → ask **before** branching, in plain text with numbered options (Codex has no structured question tool):
+- **Topic is clear** (the user said "fix X", "add Y feature") → run `bin/agent.sh branch --topic "<topic>"` automatically, continue in the printed worktree, and say one short sentence so the branch change is visible. Do not ask the user to approve routine branching.
+- **Topic is genuinely ambiguous** → ask only for the topic, using compact numbered options:
 
   ```text
-  Need a working branch first. Topic?
+  What should I call this work?
     1. <suggested slug from context>
     2. <alternative slug>
     3. Other: (name it)
   ```
 
   Wait for the reply, then branch.
+- **The user explicitly requested the protected branch** → that request is consent. Record it with `echo '{branch}' > .egregore-branch-consent`, then retry. Never create the marker merely to silence the hook.
+
+Memory, managed-repo, and runtime-state writes should bypass the project guard. If one triggers it, correct the operation target/context instead of asking for protected-branch consent.
 
 If this Codex build does not support hooks, follow the same discipline as a standing instruction: never write or commit on develop/main/master.
 
@@ -179,7 +200,22 @@ In connected mode, infrastructure credentials (Neo4j, Telegram) live on the API
 
 ## Knowledge Graph
 
-**Connected mode only.** Always use `bin/graph.sh` for Neo4j queries — never construct curl calls directly. See DEVELOPMENT.md §1 for usage examples and current schema. In local mode there is no graph; read `memory/` directly.
+**Connected mode only.** For the following intents, the named read must be the
+first retrieval action. Run the exact command with its safe default limit before
+opening files, running broad grep, inspecting `graph-op.sh`, or writing Cypher:
+
+- current work addressed to someone → `bash bin/graph-op.sh open-handoffs "<user>"`
+- unanswered questions for someone → `bash bin/graph-op.sh pending-questions "<user>"`
+- handoff → implementation → branch/PR/direction → `bash bin/graph-op.sh lineage "<topic>"`
+- company/person/topic across meetings → `bash bin/graph-op.sh meeting-history "<entity-or-topic>"`
+
+A question about calls, meetings, transcripts, demos, or how a conversation
+evolved is always `meeting-history` intent when it names a company, person, or
+topic—even when it does not ask for "current" or "live" state. Pass the shortest
+discriminative entity or topic (for example, `meeting-history "42CAP"`), not the
+whole question.
+
+Run `catalog` only when the route is unclear. Named reads return bounded stable fields and exact canonical `evidencePath` pointers; open only the returned files needed to answer. When a read reports partial coverage, use `unprojectedPaths` rather than claiming those files are represented in the graph. Use `bin/graph.sh` for unsupported Neo4j queries and never construct curl calls directly. See DEVELOPMENT.md §1 for the schema. In local mode there is no graph; read `memory/` directly.
 
 ---
 
@@ -268,13 +304,13 @@ Invoke commands from user intent — don't wait for the slash. Each command file
 **Knowledge** — `$search` `$deep-reflect` `$archive` `$note` `$add` `$meeting` `$ingest` `$scroll`
 **Identity** — `$me` (view profile or set display name)
 **Coordination** — `$ask` `$quest` `$issue` `$invite` `$delete-user` `$announce`
-**Connectors** — `$telegram-connect` (Telegram group setup)
+**Connectors** — `$telegram-connect` (Telegram group setup) `$teams-connect` (Microsoft Teams channel setup)
 **Git** — `$branch` `$commit` `$push` `$pr` `$save` `$review-pr` `$contribute`
 **Spirits** — `$summon` (persistent agent processes)
 **Infra** — `$setup` `$update` `$pull` `$env` `$infra` `$sync-repos` `$release` `$checkup`
 
 **Disambiguation:**
-- Knowledge: `$reflect` (share-ready) · `$note` (half-baked) · `$deep-reflect` (cross-reference) · `$archive` (AI patterns)
+- Knowledge: `$reflect` (share-ready) · `$note` (half-baked) · `$deep-reflect` (deep research over memory — questions AND cross-referencing) · `$archive` (AI patterns)
 - Finding things — the retrieval layers: **Grep/ls** (Claude Code / Codex file ops — always available, OSS) · **`bin/graph.sh`** (indexed relevance + relationships/status — paid). Use grep proactively; the graph deepens it when connected. (Semantic `$search` is dormant — evals showed it only pays off on massive unstructured data.)
 - Finding **something you generated** (a scroll, handoff, emissary, decision, any hosted egregore.xyz link): `bash bin/artifacts.sh find <query>`. Every generative surface already records into memory — hosted artifacts self-register into `memory/artifacts/` at publish (and the record is committed+pushed, so it's findable from *any* session, not just the one that made it); handoffs land in `memory/handoffs/`, emissaries in `memory/handoffs/outbound/`, decisions/findings/patterns in `memory/knowledge/`. The finder searches **all** of them, ranks title/topic matches above passing body mentions, tags each hit by type (`[document]`/`[handoff]`/`[emissary]`/`[decision]`), and surfaces the shareable URL. Findable by **content, not just name**, across all three layers — grep-first for OSS, folding in search + graph when available. *"bring me the artifact where I laid out our GTM plan"* → this.
 - Artifacts with questions: `$scroll` (living paper + embedded harvest, updates in place) · `$view` (static render) · `$harvest` (elicitation, no published face)
@@ -282,8 +318,8 @@ Invoke commands from user intent — don't wait for the slash. Each command file
 - Ending: `$wrap` (personal closure) · `$handoff` (notes for others) · `$save` (still working)
 - Tasks: `$todo` (personal) · `$quest` (team exploration) · `$issue` (something broken)
 - Questions: `$ask [person]` (async) · just ask (agent answers from context)
-- Ingestion: `$ingest meeting` · `$ingest user-interview` · `$ingest google` · ambiguous → ask which type
-- Connectors: `$telegram-connect` (Telegram group) · `$ingest` (bring content in)
+- Ingestion: `$ingest <file-or-folder>` (org-scoped corpus intake) · `$ingest meeting` · `$ingest user-interview` · `$ingest google` · ambiguous → ask which type
+- Connectors: `$telegram-connect` (Telegram group) · `$teams-connect` (MS Teams channel) · `$ingest` (bring content in)
 - Identity: `$me` — "who am I", "call me oz"
 - People: `$invite` (add) · `$delete-user` (remove)
 - PRs: `$pr` (create) · `$review-pr` (review)

package/runtime/codex/bin/activity-data.sh

@@ -9,7 +9,7 @@ trap "rm -rf $TMPDIR" EXIT
 
 # --- Read config ---
 API_URL=$(jq -r '.api_url // empty' "$SCRIPT_DIR/egregore.json" 2>/dev/null)
-API_KEY=$(grep '^EGREGORE_API_KEY=' "$SCRIPT_DIR/.env" 2>/dev/null | cut -d'=' -f2-)
+API_KEY="${EGREGORE_API_KEY:-$(grep '^EGREGORE_API_KEY=' "$SCRIPT_DIR/.env" 2>/dev/null | cut -d'=' -f2-)}"
 ORG=$(jq -r '.org_name // "Egregore"' "$SCRIPT_DIR/egregore.json" 2>/dev/null || echo "Egregore")
 DATE=$(date '+%b %d')