context7-skills
Original:🇺🇸 English
Translated
Use when managing Context7 CLI skills with npx ctx7 (search, install, list, remove, info).
15installs
Added on
NPX Install
npx skill4agent add narumiruna/context7-skills-skill context7-skillsTags
Translated version includes tags in frontmatterSKILL.md Content
View Translation Comparison →Context7 Skills
Purpose
This skill must directly execute Context7 CLI commands for managing skills.
Printing commands without execution is forbidden.
Scope (Authoritative & Closed)
Only the following commands are permitted (per ):
npx ctx7 skills --helpnpx ctx7 skills search|snpx ctx7 skills install|inpx ctx7 skills list|lsnpx ctx7 skills remove|rmnpx ctx7 skills info
Anything outside this list is out of scope and prohibited.
Execution Rules (Non-Negotiable)
Execution rules apply only after required permissions (if any) are granted.
- Commands MUST be executed using .
npx ctx7 - The /
skillsnamespace is mandatory.skill - Only commands listed in Scope may be executed.
- Exactly one target flag may be present at most.
- may include
install; no other install options are allowed.--all - Invalid, ambiguous, or incomplete input MUST be corrected before execution.
- No other shell commands may be executed under any circumstances.
Permission Requirements (Hard Gate)
Network-dependent commands (e.g., , , remote ) MUST NOT be executed unless the execution environment explicitly grants outbound network permission.
searchinfoinstallRequesting environment permission is not a user confirmation step and not a request for additional user input.
Permission Flow
If a command is network-dependent:
- Request outbound network permission from the execution environment / agent permission system.
- If permission is granted, execute the command.
- If permission is denied or unavailable, stop and report the limitation. Ask the user to run it locally or provide the output.
No Redundant Confirmation
If the user explicitly requested a network-dependent command and no extra input is needed, proceed to the permission request immediately without asking for additional confirmation.
Network Failures
If execution fails with a network error (e.g., , DNS, timeout), treat it as an environment/permission limitation, not a command error. Do not retry unless the environment explicitly grants permission.
fetch failedTargets (Canonical)
Valid target flags (apply only where supported by the command):
--global- —
--claude.claude/skills/ - —
--cursor.cursor/skills/ - —
--codex.codex/skills/ - —
--opencode.opencode/skills/ - —
--amp.agents/skills/ - —
--antigravity.agent/skills/
If more than one target is requested, STOP and request clarification.
Search Result Output (Mandatory)
After executing successfully:
npx ctx7 skills search|s ...- MUST display the result entries as a numbered list starting at 1.
- Each numbered item MUST preserve the entry text as-is (no paraphrase, no deduplication).
One-Run Default
- If the first run already contains any visible result entries, do not rerun. Output the visible entries immediately.
Truncation Handling (Conditional Rerun)
- Rerun the same search command up to 1 additional time only when:
- the first run shows no visible result entries, or
- the tool output is truncated () and the output contains only summary lines (e.g.,
… +N lines) without entries.Found N
- MUST NOT ask the user for confirmation to rerun if the user already requested .
search
Interactive / Anomaly Claims (Evidence Rule)
- MUST NOT claim the command became int
Install Flow From Search Selection (Mandatory)
After showing numbered search results:
- The user may reply with a number to select a skill to install.
k - On receiving a valid selection , extract from the selected entry:
kskill_name- (as shown in the entry)
repository
- If the selection is invalid (not a number or out of range), ask the user to pick a valid number.
Install Options Prompt (Numbered)
-
After a valid selection and if the user did not already specify install options, prompt the user to choose exactly one install target as a numbered list:
- (.claude/skills/)
--claude - (.cursor/skills/)
--cursor - (.codex/skills/)
--codex - (.opencode/skills/)
--opencode - (.agents/skills/)
--amp - (.agent/skills/)
--antigravity - (global)
--global
-
The user responds with a number. Map
tto the corresponding target flag.t -
MUST NOT be used when installing a single selected skill.
--all
Permission Gate
- If the install requires network access (remote repository), request outbound network permission before running install.
Execute Install
- Execute exactly:
npx ctx7 skills install|i <repository> <skill_name> <target_flag>
- After execution, display the raw CLI output as-is.
Quick Reference
| Intent | Executed Command |
|---|---|
| Search | `npx ctx7 skills search |
| Install | `npx ctx7 skills install |
| List | `npx ctx7 skills list |
| Remove | `npx ctx7 skills remove |
| Repo info | |
Examples (Executed)
bash
npx ctx7 skills search pytorch vision
npx ctx7 skills install /anthropics/skills uv --claude
npx ctx7 skills install /anthropics/skills --all --codex
npx ctx7 skills list --cursor
npx ctx7 skills info /anthropics/skillsHard Errors (Block Execution)
- Missing /
skillsskill - Using commands outside Scope
- Emitting commands without execution
- More than one target flag
- Mixing targets with unrelated flags
- Attempting to run non-Context7 shell commands
- Executing network-dependent commands without explicit permission
Notes (Behavioral)
- is mandatory for
<repository>andinstall.info - is optional; omission triggers interactive selection unless
[skill]is used.--all - does not support filtering.
list - Network failures (e.g., ) indicate environment/permission limits, not a command syntax issue.
fetch failed