BC codexbox
OpenAI Codex CLI running inside an aicodebox container, put on the network. Exposes seven ways in — interactive shell, one-shot exec, an HTTP REST API (workspace file ops, sync/async prompt runs with run-id polling), an OpenAI-compatible /openai/v1/chat/completions endpoint (streaming, client-executed tools/tool_choice, response_format/JSON-schema), an MCP server (streamable HTTP, mounted at /mcp in API mode or as a sidecar), a Telegram bot, and a cron scheduler that fires codex on a schedule. Auth is bearer-token per surface (CODEXBOX_API_MODE_TOKEN, CODEXBOX_MCP_MODE_TOKEN) plus codex's own OpenAI API-key or ChatGPT-subscription login. Use when the user wants to run OpenAI Codex programmatically over HTTP/MCP/Telegram/cron instead of only in a local terminal, or wants an OpenAI-compatible endpoint backed by Codex.
OpenAI Codex CLI running inside an aicodebox container, put on the network.
As a process C 57/100 · Has gaps — weak spots: result and completion, inputs and preconditions, running it twice
What is at stake
Medium-severity findings: the skill is probably honest, but read what alarmed the scanner.
Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.
The instructions or scripts send data out: environment variables, keys, file contents, chat history. You may never notice, because the agent performs the upload as "part of the task".
If the upload is not needed for the task, remove it: catalogs flag such skills and delist them. If it is needed, name the destination explicitly, say what leaves the machine, and give the user a switch.
How to improve
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 5
✓ No critical or high findings
Medium and low: 5
-
medium Exfiltration
net-credential-useSKILL.md:126Credential used in a network call (verify the destination is the intended service)curl -sS -X PUT --oa…rer "$CODEXBOX_API_MODE_TOKEN" \
-
medium Exfiltration
net-credential-useSKILL.md:129Credential used in a network call (verify the destination is the intended service)curl -sS --oa…rer "$CODEXBOX_API_MODE_TOKEN" \
-
medium Exfiltration
net-credential-useSKILL.md:301Credential used in a network call (verify the destination is the intended service)run_id=$(curl -s http://loca…080/run -H "Authorization: Bearer $CODEXBOX_TOKEN" \
-
medium Exfiltration
net-credential-useSKILL.md:305Credential used in a network call (verify the destination is the intended service)curl -s "http://loca…080/run/result?runId=…" -H "Authorization: Bearer $CODEXBOX_TOKEN" | jq
-
medium Exfiltration
net-credential-useSKILL.md:311Credential used in a network call (verify the destination is the intended service)curl -s -X DELETE "http://loca…080/run/$run_id" -H "Authorization: Bearer $CODEXBOX_TOKEN"
Files scanned: 3. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- note
frontmatter-keyunknown frontmatter key "homepage"
Process rating: all ten parameters 57/100
- 0Result and completion. Does not say what the result is
- 0Inputs and preconditions. Does not say what the process needs to start
- 30Running it twice. 25 mutating operations with no state check
- 60Tools and files. Uses tools (bash, web) that frontmatter does not declare
- 70When it triggers. States when to use, but not when not to
- 70Execution cost. Instruction body is 4070 tokens
- 85Steps. 22 steps, 1 vague phrases
- 100Failures and branches. 1 branches, has a failure section
- 100Consistency. Name and required fields are in place
- 100Progress reporting. Reports progress
- medium Safety rules and hard prohibitions inside a skill: they belong in the system prompt, here they protect nothing
- low 12 top-level sections: this looks like several domains in one skill
- low The response is described with custom markup (6 tags): a typed call is more reliable
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- +4Description does not say when NOT to use the skill (false activations)
- +3Description length 827: 120–800 characters recommended
- +3Output format is not stated: the model decides each time
- -2localhost URLs: will not work for another user
- +1No license
- +2Single-language instructions
- +4Structure: 13 headings
- +3Step-by-step instructions: 22 items
- +4Has examples (19 code blocks)
- +4Reference files are cited in the instructions (1 of 1)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 82.