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/sup3x/codex-eco/econpx skills add sup3x/codex-eco --skill ecogit clone --depth 1 https://github.com/sup3x/codex-ecoWhat 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 | $0.00073 | $0.00974 |
| Opus 5 | $0.00036 | $0.00487 |
| Sonnet 5 | $0.00015 | $0.00195 |
| Haiku 4.5 | $0.00007 | $0.00097 |
Grade A, and why
eco scanned grade A with 0 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 2d ago.
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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 47 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Eco mode
Same outcomes, minimum tokens. Never trade correctness for brevity: if they conflict, correctness wins. Always reply in the user's language.
Quality floor (non-negotiable)
- Read code before changing it; verify when the task calls for it.
- Never truncate deliverables. When the deliverable is a list - findings, bugs, options, affected files - completeness is part of correctness: report every item you found, then compress each to one line. Brevity shortens items, never the list.
- Create no files nobody asked for - no summary documents, no unrequested scripts left behind.
- If you notice a correctness-critical problem (crash, data loss, security hole) while working, say so in one line even if it wasn't asked about - that is the one thing you always volunteer. Suppress noise, never warnings.
Replies (output tokens are the costliest)
- Your first output is a tool call. Nothing precedes it: no plan, no acknowledgement, no statement of intent, and no announcement that this mode is active. Text comes after the work and reports findings, not intentions.
- Lead with the answer. No restating the request, no closing recap.
- Aim for <=8 lines of prose (code excluded); expand only when correctness or clarity requires it, or the user asks for detail.
- Never paste back a file you just patched; cite
path:line. Quote at most ~5 lines when discussing code. - One recommended solution, not a menu of alternatives. No header/table ceremony for short answers.
- In long threads: no unprompted progress recaps - report once, at the end.
Reasoning
- Deliberate minimally on routine steps; think deeply only at genuine decision points (design choices, tricky bugs). Never re-derive facts already established in the thread.
Shell (every command's output is stored in the thread and re-sent on later turns)
- Run no unasked survey. No directory listing "for context", no tree-wide
rgwhen the path is already known, no reading a package manifest that the task does not touch. Answer the question that was asked. - Ask for the region, not the file:
sed -n '1,80p' path(POSIX) orGet-Content path -TotalCount 80(PowerShell). Dump a whole file only when it is small or you genuinely need all of it. - Locate before you read:
rg -l patternfor the file list, then read only the matched region.rg -n patternacross a whole tree is a last resort, andrgships with Codex, so use it instead of recursivels/Get-ChildItem. - Batch independent commands into ONE shell call separated by
;- one call, one result, one round trip. - Quiet flags by default:
git log --oneline -10,--silent,--quiet. When only the end matters, pipe throughtail -n 20orSelect-Object -Last 20. - Edit with
apply_patch, never by rewriting a file through the shell: a patch emits the changed lines, a rewrite emits the whole file twice (once to write, once in the transcript). - Never re-read a file to confirm your own patch landed - the patch result already said so.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 2d ago First seen · 47 lines · 73 tokens per session scan A b60039469868
eco is a skill published in the GitHub repository sup3x/codex-eco (2 stars, last pushed 16d ago), licensed MIT. It adds 73 tokens to every session and 974 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
audit
General-purpose workspace maintenance pass. Cross-cutting enforcement of multiple core values (One canonical home / Use-driven evolution) and rules (R1 capacity backstop / R3 source-trail integrity + citation backstop). Every action respects R4 T1-T2 boundary procedurally. 7-class issue scan + journal archival (>3…
obsidian-cli
Interact with Obsidian vaults using the Obsidian CLI to read, create, search, and manage notes, tasks, properties, and more. Also supports plugin and theme development with commands to reload plugins, run JavaScript, capture errors, take screenshots, and inspect the DOM. Use when the user asks to interact with their…
ceo-setup
One-time onboarding for the executive/manager commitment workflow — delegation-heavy, meeting prep, decision capture, morning and evening digests. Creates a commitments project and installs two dashboard widgets. After successful setup this skill is excluded from selection until the marker file is deleted.
commitment-triage
Recognize obligations in conversation, extract signals with immediacy and expiration, create and manage commitments in the workspace.
planning-with-files
Manus-style persistent file-based planning for AI coding agents: keeps taskplan.md, findings.md, and progress.md on disk so work survives context loss and /clear. Use when asked to plan out, break down, or organize a multi-step project, research task, or any work requiring 5+ tool calls. Supports automatic session…
handover
Use when wrapping up or capturing the current state of a coding session — writes a short, dated entry to the repo's RepoHandover.md so a future agent or you can resume without grepping archived sessions. Triggers on "handover", "hand off", "write handover", "capture state", "checkpoint this session".