How do I install the Proof Grammar Island skill?

Run npx skillmds@latest add island-dev-crew/proof-grammar-island in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Proof Grammar Island safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Proof Grammar Island?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Proof Grammar Island free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Proof Grammar Island?

Island-Dev-Crew (@island-dev-crew) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Proof Grammar Island by island-dev-crew

Name: Proof Grammar Island
Author: Island-Dev-Crew

Proof Grammar Island

A fixture whose proof block carries every shape of the proof grammar at once - a bare island-relative script, a distant exit-report line, an off-allowlist command, a usage template, and a candidate with no exit code. Never distributed; exists so the extractor's classification is a captured run rather than a claim. Trigger phrases - "proof grammar fixture", "pending red test".

File contents

Proof grammar island

One block, five classifications. The first four shapes were each silently dropped or silently misread before the grammar was written down (C4).

enforced: verify-proofs.py runs the two proofs, and --strict goes red on the PENDING.
advisory: nothing else here is real.

scripts/probe.sh dirty   # exit 1
$ scripts/probe.sh clean
| PASS nothing dirty
$ echo $?   # → 0
python3 -c "import sys; sys.exit(0)"
scripts/probe.sh <mode>
true   # exit 0

The five, in order: a bare island-relative script with an inline code; the same script whose code sits on a report line two lines below its output; a candidate carrying no code at all (PENDING); a usage template with a <placeholder> (TEMPLATE); an off-allowlist leading token that still states a code (SKIPPED).

Two proofs run here, so the PENDING is not the only thing --strict has to weigh. pending-only/ is the island where it is, and where the two answers compete.

unrecognized-command/ captures the report-binding boundary: an off-allowlist command between a candidate and an exit report is SKIPPED; the earlier candidate stays PENDING and cannot borrow that later status.

Two narrower grammar edges sit beside it. bash-builtin/ proves an off-allowlist Bash builtin is the same command boundary, and empty-assignment/ proves an empty ordinary binding clears stale block context rather than disappearing.

host-independent-command/ removes the last lookup guess: a syntactically command-shaped line owns its position whether or not that name exists on the review host. Ambiguous shell-looking output stops report search too; | marks explicit output, and inline exit annotations are the other unambiguous form.

malformed-command/ extends that boundary to rows a shell lexer rejects, while export-boundary/ proves a replayable export still owns the status it produces. report-compound/ requires a pure report to end where the report grammar ends. comment-lexing/ watches hashes protected by ANSI-C quoting or escaped whitespace, and continuation/ watches Bash's zero-byte backslash-newline removal. The parser self-check below keeps encoded forbidden names entirely out of a subprocess while proving hex, octal, and Unicode spellings are refused in direct commands and all four substitution forms.

The host-independence and lexer claims are executable rather than inferred from this prose:

python3 check-host-independent.py   # exit 0
python3 check-parser-regressions.py  # exit 0

---name: proof-grammar-islanddescription: A fixture whose proof block carries every shape of the proof grammar at once - a bare island-relative script, a distant exit-report line, an off-allowlist command, a usage template, and a candidate with no exit code. Never distributed; exists so the extractor's classification is a captured run rather than a claim. Trigger phrases - "proof grammar fixture", "pending red test".---# Proof grammar islandOne block, five classifications. The first four shapes were each silently dropped orsilently misread before the grammar was written down (C4).- `enforced`: verify-proofs.py runs the two proofs, and `--strict` goes red on the PENDING.- `advisory`: nothing else here is real.```bashscripts/probe.sh dirty # exit 1$ scripts/probe.sh clean| PASS nothing dirty$ echo $? # → 0python3 -c "import sys; sys.exit(0)"scripts/probe.sh <mode>true # exit 0```The five, in order: a bare island-relative script with an inline code; the same scriptwhose code sits on a report line two lines below its output; a candidate carrying no codeat all (PENDING); a usage template with a `<placeholder>` (TEMPLATE); an off-allowlistleading token that still states a code (SKIPPED).Two proofs run here, so the PENDING is not the only thing `--strict` has to weigh.[`pending-only/`](pending-only/SKILL.md) is the island where it is, and where the two answerscompete.[`unrecognized-command/`](unrecognized-command/SKILL.md) captures the report-binding boundary:an off-allowlist command between a candidate and an exit report is `SKIPPED`; the earliercandidate stays `PENDING` and cannot borrow that later status.Two narrower grammar edges sit beside it. [`bash-builtin/`](bash-builtin/SKILL.md) proves anoff-allowlist Bash builtin is the same command boundary, and[`empty-assignment/`](empty-assignment/SKILL.md) proves an empty ordinary binding clears staleblock context rather than disappearing.[`host-independent-command/`](host-independent-command/SKILL.md) removes the last lookup guess:a syntactically command-shaped line owns its position whether or not that name exists on thereview host. Ambiguous shell-looking output stops report search too; `| ` marks explicit output,and inline exit annotations are the other unambiguous form.[`malformed-command/`](malformed-command/SKILL.md) extends that boundary to rows a shell lexerrejects, while [`export-boundary/`](export-boundary/SKILL.md) proves a replayable export still ownsthe status it produces. [`report-compound/`](report-compound/SKILL.md) requires a pure report toend where the report grammar ends. [`comment-lexing/`](comment-lexing/SKILL.md) watches hashesprotected by ANSI-C quoting or escaped whitespace, and[`continuation/`](continuation/SKILL.md) watches Bash's zero-byte backslash-newline removal. Theparser self-check below keeps encoded forbidden names entirely out of a subprocess while provinghex, octal, and Unicode spellings are refused in direct commands and all four substitution forms.The host-independence and lexer claims are executable rather than inferred from this prose:```bashpython3 check-host-independent.py # exit 0python3 check-parser-regressions.py # exit 0```

Frequently asked questions