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 --help):

npx ctx7 skills search|s
npx ctx7 skills install|i
npx ctx7 skills list|ls
npx ctx7 skills remove|rm
npx 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 skills / skill namespace is mandatory.
Only commands listed in Scope may be executed.
Exactly one target flag may be present at most.
install may include --all; no other install options are allowed.
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., search, info, remote install) MUST NOT be executed unless the execution environment explicitly grants outbound network permission.

Requesting 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., fetch failed, DNS, timeout), treat it as an environment/permission limitation, not a command error. Do not retry unless the environment explicitly grants permission.

Targets (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 npx ctx7 skills search|s ... successfully:

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 (… +N lines) and the output contains only summary lines (e.g., Found N) without entries.
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 k to select a skill to install.
On receiving a valid selection k, extract from the selected entry:
- skill_name
- repository (as shown in the entry)
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:
1. --claude (.claude/skills/)
2. --cursor (.cursor/skills/)
3. --codex (.codex/skills/)
4. --opencode (.opencode/skills/)
5. --amp (.agents/skills/)
6. --antigravity (.agent/skills/)
7. --global (global)
The user responds with a number t. Map t to the corresponding target flag.
--all MUST NOT be used when installing a single selected skill.

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	`npx ctx7 skills info <repository>`

Examples (Executed)

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/skills

Hard Errors (Block Execution)

Missing skills / skill
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)

<repository> is mandatory for install and info.
[skill] is optional; omission triggers interactive selection unless --all is used.
list does not support filtering.
Network failures (e.g., fetch failed) indicate environment/permission limits, not a command syntax issue.

context7-skills