timerloggedout-spec · Sep 19, 2026 · Sep 19, 2026 · Sep 19, 2026
diff --git a/.agents/skills/github-issue-pr-graph/SKILL.md b/.agents/skills/github-issue-pr-graph/SKILL.md
@@ -0,0 +1,61 @@
+---
+name: github-issue-pr-graph
+description: Evaluate, refine, correlate, and assign GitHub issues and pull requests. Use for sub-issues, labels, projects, message threads, review threads, commit comments, mention-graph reconstruction, and PR status plus discussion management. Load with evidence-led-monorepo-ops, adaptive-wait, and context-relationship-graph.
+---
+
+# Skill: github-issue-pr-graph
+
+**Canonical:** `.agents/skills/github-issue-pr-graph/SKILL.md` on **master**.
+Companion graph substrate: `.agents/skills/context-relationship-graph/SKILL.md`.
+
+**Load with:** `evidence-led-monorepo-ops` · `adaptive-wait` · `review-loop`.
+
+## Intent
+
+Operate the GitHub social graph for this monorepo without comment-storms.
+
+| Verb | Meaning |
+|------|---------|
+| Evaluate | Read issue/PR + reviews + checks + linked items before acting |
+| Refine | Narrow title/body/labels; extract a sub-issue instead of bloating one thread |
+| Correlate | Map issue ↔ PR ↔ commit ↔ review thread ↔ project card ↔ mention |
+| Assign | Set assignee/reviewer only when the person or bot is the intended actor |
+| Sub-issues | Split scope; keep parent as tracker |
+| Labels | Use existing repo labels; do not invent taxonomy in comments |
+| Projects | Move cards; do not duplicate status in comments |
+| Threads | Reply in the existing review/issue thread; one durable fact per action |
+
+## Mention graph (test / invoke)
+
+Official monikers live in `docs/ops/AGENT-MONIKERS.md`.
+
+| Channel | Call |
+|---------|------|
+| DeepSeek CI | `@deepseek` / `@deepseek-ci` / `@deepCore` or labels `deepseek-ci`, `deepseek`, `deepCore` |
+| CodeRabbit | `@coderabbitai` |
+| Jules | `@jules` |
+| Gemini CLI | `@gemini-cli` |
+
+DeepSeek web-wrapper needs a live `DEEPSEEK_COOKIES` / session secret. Stale tokens fail `create_chat_session` with `40003`. That is expected after rotation — do not re-derive cookies from exposed profiles (`docs/CREDENTIAL-EXPOSURE.md`).
+
+## Operating loop
+
+1. Recon open issues + PRs (state, base SHA vs live master, mergeable, reviews).
+2. Correlate with `context-relationship-graph` when history matters.
+3. Act once — label, assign, open sub-issue, or push a scoped PR.
+4. Record durable facts in the PR/issue body or `docs/ops` — not a new comment for every poll.
+5. Dual-gate before merge (`repo_gate.py` + `termux_smoke.py`).
+6. WAIT on CI; stay busy on disjoint work.
+
+## Neighbor-safe rules
+
+- Do not re-open closed help-wanted claims.
+- Do not merge dirty/behind record PRs; extract onto live master.
+- Do not force-merge mega-PRs.
+- CodeRabbit full-repo review is requested with `@coderabbitai` on the PR plus org app install (Operator).
+
+## Closeout
+
+State evaluated items, correlations (issue/PR/SHA), labels/projects touched, threads resolved vs left WAIT, and the next dual-gate SHA.
+
+Agent-Identity: Grok (Administrator)
diff --git a/AGENTS.md b/AGENTS.md
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -1,6 +1,7 @@
 # CLAUDE.md
 
-**Primary agent entry for this repository.** Load this first. Do not start from root `AGENTS.md` (Linguist/CedrLang stub only).
+**Primary and only root agent entry for this repository.** Load this first.
+Root `AGENTS.md` is **deleted** (2026-09-19). Linguist / CedrLang practice surfaces live under docs and generated artifacts — not a second governance file.
 
 ## Governance sequence (read in order)
 
@@ -59,7 +60,10 @@ Inventory: [`docs/ops/SKILLS-INVENTORY.md`](docs/ops/SKILLS-INVENTORY.md)
 | Skills + adaptive wait | `docs/ops/SKILLS-INVENTORY.md` |
 | Session SSOT | `docs/schemas/session-ssot.md` |
 | Provider caps | `docs/schemas/provider-capabilities.md` |
-| Linguist / CedrLang surface only | `AGENTS.md` |
+| Issue/PR graph ops | `.agents/skills/github-issue-pr-graph/SKILL.md` |
+| Context relationship graph | `.agents/skills/context-relationship-graph/SKILL.md` |
+| Agent monikers / @-invoke | `docs/ops/AGENT-MONIKERS.md` |
+| DeepSeek CI invoke | `docs/ops/DEEPSEEK-CI.md` |
 
 ## ICM + RinDig reference pins (smods)
 

diff --git a/SKILLS.md b/SKILLS.md
@@ -9,6 +9,7 @@
 | Adaptive wait / feedback | `.agents/skills/adaptive-feedback-cycle/SKILL.md` |
 | Admin ops | `.agents/skills/evidence-led-monorepo-ops/SKILL.md` |
 | External contribute (help-wanted) | `.agents/skills/help-wanted-lane/SKILL.md` |
+| Issue / PR graph (evaluate–assign–threads) | `.agents/skills/github-issue-pr-graph/SKILL.md` |
 | Production WAIT → VALIDATE | `.github/skills/production-reconciliation/SKILL.md` |
 
 ## Layout (both trees are first-class)
@@ -22,13 +23,16 @@ SKILLS.md                           # this file — short collaborator entry
 
 Every skill directory must contain a `SKILL.md`. Inventory lists **all** of them so collaborators do not need private mirrors.
 
+Root `AGENTS.md` is removed. Start at `CLAUDE.md`.
+
 ## Role quick-load
 
 | Role | Load first |
 |------|------------|
 | Collaborator | `adaptive-feedback-cycle` → dual-gate |
 | Admin / Grok | `evidence-led-monorepo-ops` + `adaptive-wait` |
 | Oversight / external PR | `help-wanted-lane` |
+| Issue/PR graph + threads | `github-issue-pr-graph` |
 | Evaluation / DOE | `multivariate-doe` + `blind-agent-evaluation` |
 
 Do not keep skill policy only in local `.grok/skills/` mirrors — **master is SSOT**.

diff --git a/docs/ops/SKILLS-INVENTORY.md b/docs/ops/SKILLS-INVENTORY.md
@@ -4,6 +4,8 @@
 **Primary agent entry:** [`CLAUDE.md`](../../CLAUDE.md)  
 **Collaborator short entry:** [`SKILLS.md`](../../SKILLS.md)
 
+Root `AGENTS.md` is **removed**. Do not recreate it as a second governance file.
+
 ## Collaborator access (parity)
 
 All skills below are **in-repo** under `.agents/skills/` or `.github/skills/`.  
@@ -19,9 +21,10 @@ Local agent mirrors (e.g. `.grok/skills/`) are convenience only — **not** a se
 
 | Role | Load first | Then |
 |------|------------|------|
-| **Admin / Grok Administrator** | `evidence-led-monorepo-ops` + `adaptive-wait` | `review-loop` · `help-wanted-lane` · `approxination-lane` · `production-reconciliation` |
+| **Admin / Grok Administrator** | `evidence-led-monorepo-ops` + `adaptive-wait` | `review-loop` · `help-wanted-lane` · `github-issue-pr-graph` · `approxination-lane` · `production-reconciliation` |
 | **Collaborator** | `adaptive-feedback-cycle` | dual-gate · `find-skills` |
 | **Oversight / external contrib** | `help-wanted-lane` | evidence-led + adaptive-wait · living status board |
+| **Issue/PR graph** | `github-issue-pr-graph` | `context-relationship-graph` · `review-loop` |
 | **Evaluation / DOE** | `multivariate-doe` + `blind-agent-evaluation` | `approxination-lane` · `mvt-experiment` |
 | **CI / production recon** | `production-reconciliation` | `action-effectiveness-ledger` · `workflow-orchestration` |
 
@@ -37,6 +40,7 @@ Local agent mirrors (e.g. `.grok/skills/`) are convenience only — **not** a se
 | evidence-led-monorepo-ops | `.agents/skills/evidence-led-monorepo-ops/SKILL.md` |
 | find-skills | `.agents/skills/find-skills/SKILL.md` |
 | gemini-performance-psychology | `.agents/skills/gemini-performance-psychology/SKILL.md` |
+| github-issue-pr-graph | `.agents/skills/github-issue-pr-graph/SKILL.md` |
 | help-wanted-lane | `.agents/skills/help-wanted-lane/SKILL.md` |
 | multivariate-doe | `.agents/skills/multivariate-doe/SKILL.md` |
 | review-loop | `.agents/skills/review-loop/SKILL.md` |
@@ -74,6 +78,8 @@ Local agent mirrors (e.g. `.grok/skills/`) are convenience only — **not** a se
 | Model rotation | [`docs/schemas/model-rotation.yaml`](../schemas/model-rotation.yaml) |
 | Provider caps | [`docs/schemas/provider-capabilities.md`](../schemas/provider-capabilities.md) |
 | Approxination ops | [`APPROXINATION-LANE.md`](APPROXINATION-LANE.md) |
+| DeepSeek invoke | [`DEEPSEEK-CI.md`](DEEPSEEK-CI.md) |
+| Agent monikers | [`AGENT-MONIKERS.md`](AGENT-MONIKERS.md) |
 
 ## Cycle snapshot (2026-09-19)
 
@@ -82,5 +88,6 @@ Local agent mirrors (e.g. `.grok/skills/`) are convenience only — **not** a se
 - Proven PRIMARY: DioNanos/codex-termux PR #27 (do **not** re-claim closed #14); Haven PR #657.
 - Maintainer routing (codex-termux): parity only; logic → upstream; features → [codex-vl](https://github.com/DioNanos/codex-vl).
 - Dual-gate still required for monorepo merges. External PRs follow target norms.
+- **This cycle:** add `github-issue-pr-graph`; fold and delete root `AGENTS.md`.
 
 BIUDL. Agent-Identity: Grok (Administrator)