Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/optave/ops-codegraph-tool/overseenpx skills add optave/ops-codegraph-tool --skill overseegit clone --depth 1 https://github.com/optave/ops-codegraph-toolWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00143 | $0.16726 |
| Opus 5 | $0.00072 | $0.08363 |
| Sonnet 5 | $0.00029 | $0.03345 |
| Haiku 4.5 | $0.00014 | $0.01673 |
Grade D, and why
oversee scanned grade D with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
3. **Auto-accept ("auto mode") — CONFIRM, not detect.** There is no env var or settings field that reports the **live** permission mode: it is runtime state toggled with shift+Tab. It can only be confirmed, never detecte Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
**Cleanup.** State is safe to re-read and safe to delete: a fresh invocation re-derives everything it needs from GitHub. Remove it with `rm -rf .codegraph/oversee` once the execute PR is merged. Leaving it costs nothing. How it starts
The opening of the file, as written. The whole thing — 947 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/oversee — human-gated, single-issue delivery for codegraph
/oversee takes one GitHub issue from "nobody has planned this" to "a verified, reviewer-clean PR waiting for a human merge", with exactly one human decision in the middle: does this plan get built? Everything else — planning, adversarial plan review, reviewer sweeps, the build, independent verification, review convergence — runs automatically.
It is the human-in-the-loop counterpart to /fixer. The two are complementary, not interchangeable:
/fixer |
/oversee |
|
|---|---|---|
| Scope | the whole qualifying backlog, in batches | exactly one issue |
| Plan | none — solve directly | a committed plan doc, reviewed and human-approved before any code |
| Human decision | none until the final report | one: tick the approval box |
| Merge | merges each PR itself | never merges — the human does |
Use /oversee for work where getting the approach wrong is expensive: anything touching the dual-engine boundary, resolution semantics, the language registry, a public API, or an architectural decision. Use /fixer for a backlog of self-contained fixes.
You are split in two:
- You (this skill) are the gating brain. You preflight the session config, route the run, run a readiness check, drive the human approval gate, verify its provenance, validate plan freshness, and reconcile the reviewer after execution. You never write product code.
oversee-dispatchis the execution engine. You hand it one task plus aphase, and it runs that half of the pipeline.
/oversee #<issue> ─preflight─▶ sync main ─▶ readiness ─▶ PLAN ─▶ critic (advisory, revise-in-place)
─▶ plan-sweep ─▶ [Plan] PR + ✋ approval box ─▶ STOP
│ (a human ticks the box)
/oversee #<plan-PR> ─preflight─▶ sync main ─▶ box ticked? ─▶ provenance? ─▶ freshness gate (Sonnet subagent)
├─FRESH──▶ EXECUTE (pinned to approved SHA) ─▶ VERIFY ─▶ execute-sweep ─▶ reconcile ─▶ STOP (human merges)
├─NARROW─▶ patch the plan in place ─▶ re-stamp the gate ─▶ STOP (needs a fresh tick)
└─STALE──▶ carry-forward ─▶ close the [Plan] PR ─▶ re-PLAN ─▶ STOP (needs a fresh tick)
Two things make this worth running instead of just asking an agent to fix an issue:
- A preflight config gate runs before any compute is spent.
- The human is the plan gate, and the gate cannot be forged. The critic is advisory — its verdict is printed next to the checkbox, it never auto-advances. EXECUTE runs only after a human ticks the box, and only when a commit status the PR body's author cannot write proves the gate was installed by an
/overseerun for that exact plan head.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 947 lines · 143 tokens per session scan D cacc7e17d925
oversee is a skill published in the GitHub repository optave/ops-codegraph-tool (92 stars, last pushed 3d ago), licensed Apache-2.0. It adds 143 tokens to every session and 16,726 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it D with 2 findings (reads agent configuration directories, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
codespaces
Build and query architecture-aware code search, dependency graphs, call flows, impact analysis, and belief maps for Python, TypeScript, TSX, Rust, C#, Java, Go, Ruby, and Ruby on Rails repositories. Use before non-trivial code changes to find module boundaries, blast radius, architecture violations, and the minimal…
security-triage
Turn a large codebase into a short, ranked reading list of security-relevant code paths worth a human researcher's time, using the flyto-indexer MCP tools. Use when asked to find, prioritize, or triage potential vulnerabilities / taint flows / attack surface in a repository indexed (or indexable) by flyto-indexer …
devlens
Understand a codebase with the DevLens MCP — TypeScript, JavaScript, Python, Go, Rust, or Java (incl. React/Next.js/Node, FastAPI/Flask/Django, Spring Boot, Gin/Echo/chi/net-http, axum/actix/rocket). Query a precomputed graph of nodes (components, hooks, functions, classes, methods, structs, traits, routes) and typed…
archon
Multi-language dependency impact analysis for Java, JS/TS, and Python codebases. Answers "if I change this, what breaks?" Run /archon diff before refactoring, /archon analyze for full dependency maps.
codebase-intelligence
Query the codebase-intelligence CLI to understand TypeScript architecture, dependencies, blast radius, and risk before reading files. Use for any "how is this structured", "what breaks if I change X", "where is the complexity" question.
repomix
Pack and analyze codebases into AI-friendly single files using Repomix. Use when the user wants to explore repositories, analyze code structure, find patterns, check token counts, or prepare codebase context for AI analysis. Supports both local directories and remote GitHub repositories.