Test results

What Morphness is tested with, what came out, and when. Release 2026-10-20.4 + later commits, commit 98324c8, run 2026-10-20 17:14 UTC.

This page is written by scripts/trust-report.sh from a real run, never by hand. A tool that fails to start is shown as "did not run", not as a pass. Every warning, error or finding gets a verdict in tests/triage.jsonl (fixed, false positive or accepted, with the reason); one without a verdict is shown as untriaged. "Tested with" is all this page claims. It is not a certificate that Morphness is secure.

Morphness's own checks

Anyone who clones the repository can run these.

Check What it looks at Result Run

Check	What it looks at	Result	Run
Kit self-test	Renders the template as /morph-init would for sample projects, runs its own consistency checks, replays every guard case and runs the template scripts. `STRICT=1 GUARD_FULL=1 scripts/kit-selftest.sh`	4 of 4 sample projects pass; 616 guard cases replayed on each (2464 replays; per project 203 allowed, 272 asked, 141 denied, all as expected) Sample projects: go-service, monorepo-apps, nextjs-supabase, python-cli. Raw output: kit-selftest.txt	2026-10-20 17:14 UTC
Update notes	Every kit change has an update note, and every note is well formed. `scripts/check-updates.sh`	85 update note(s) checked, 0 problem(s) Raw output: check-updates.txt	2026-10-20 17:14 UTC
Lesson catalog	Every lesson file follows the catalog format. `scripts/check-lessons.sh`	112 lesson file(s) checked, 0 problem(s) Raw output: check-lessons.txt	2026-10-20 17:14 UTC

Kit self-test

Renders the template as /morph-init would for sample projects, runs its own consistency checks, replays every guard case and runs the template scripts.
STRICT=1 GUARD_FULL=1 scripts/kit-selftest.sh

4 of 4 sample projects pass; 616 guard cases replayed on each (2464 replays; per project 203 allowed, 272 asked, 141 denied, all as expected)

Sample projects: go-service, monorepo-apps, nextjs-supabase, python-cli.

Raw output: kit-selftest.txt

2026-10-20 17:14 UTC

Update notes

Every kit change has an update note, and every note is well formed.
scripts/check-updates.sh

85 update note(s) checked, 0 problem(s)

Raw output: check-updates.txt

2026-10-20 17:14 UTC

Lesson catalog

Every lesson file follows the catalog format.
scripts/check-lessons.sh

112 lesson file(s) checked, 0 problem(s)

Raw output: check-lessons.txt

2026-10-20 17:14 UTC

Public tools

Pinned versions, run locally with no API key. Nothing is uploaded to the tools' vendors.

Check	What it looks at	Result	Run
agnix 0.59.0	Config correctness of CLAUDE.md, AGENTS.md, rules, agents, settings, hooks and skills, on a rendered sample install and on the kit's skills. `npx agnix@0.59.0 --format json <path>`	3 finding(s), all judged (8 informational) 3 finding(s) and how each was judged `CDX-AG-005` install/AGENTS.md:33 AGENTS.md references missing file `QA/qa-result.md` false-positive: QA/qa-result.md is created by the first QA run, so it does not exist in a fresh install. `PE-004` install/AGENTS.md:35 Ambiguous term 'usually' creates unclear instruction accepted: 'usually' is a deliberate hedge in a role description; the rule is a style check. `XML-001` install/harness/templates/steps.md:11 Unclosed XML tag '<Service>' false-positive: <Service> and <tool> are placeholders in an example sentence, not XML. 66 files checked. Errors and warnings need a verdict; info notes are only counted. Raw output: agnix-install.json · agnix-skills.json	2026-10-20 17:14 UTC
ShellCheck 0.11.0	The kit's bash scripts and the scripts installed into projects, including the guard hook. `shellcheck -x -f json $(git ls-files '*.sh')`	8 finding(s), all judged (9 informational) 8 finding(s) and how each was judged `SC2010` template/scripts/prune-counter.sh:85 Don't use ls \| grep. Use a glob or a for loop with a condition to allow non-alphanumeric filenames. accepted: Lists the kit's own prune files, whose names are dated slugs; style warning. `SC1112` template/scripts/guard.sh:619 This is a unicode quote. Delete and retype it (or ignore/doublequote for literal). false-positive: A typographic apostrophe inside the text of a user-facing message, on purpose. `SC1112` template/scripts/guard.sh:639 This is a unicode quote. Delete and retype it (or ignore/doublequote for literal). false-positive: A typographic apostrophe inside the text of a user-facing message, on purpose. `SC1112` template/scripts/guard.sh:1183 This is a unicode quote. Delete and retype it (or ignore/doublequote for literal). false-positive: A typographic apostrophe inside the text of a user-facing message, on purpose. `SC1112` template/scripts/guard.sh:1236 This is a unicode quote. Delete and retype it (or ignore/doublequote for literal). false-positive: A typographic apostrophe inside the text of a user-facing message, on purpose. `SC2155` template/scripts/evidence.sh:22 Declare and assign separately to avoid masking return values. accepted: Declare-and-assign of config values that cannot fail in a way the script needs to see; style warning. `SC2155` template/scripts/evidence.sh:23 Declare and assign separately to avoid masking return values. accepted: Declare-and-assign of config values that cannot fail in a way the script needs to see; style warning. `SC2155` template/scripts/evidence.sh:24 Declare and assign separately to avoid masking return values. accepted: Declare-and-assign of config values that cannot fail in a way the script needs to see; style warning. 13 scripts. Errors and warnings need a verdict; style notes and info are only counted. Raw output: shellcheck.json	2026-10-20 17:14 UTC
gitleaks 8.28.0	Secrets in the files and in the whole git history (matches are redacted in the log). `gitleaks detect --redact`	0 findings 95 commits scanned. Raw output: gitleaks.json	2026-10-20 17:14 UTC
SkillSpector 2.12.0	Prompt injection, exfiltration and tool poisoning patterns in the five /morph-* skills. Static mode only, no model is called. `skillspector scan skills/<name> --no-llm`	0 findings Risk scores (0 is best): morph-doctor 0/100, morph-init 0/100, morph-module 0/100, morph-share 0/100, morph-update 0/100. Raw output: skillspector-morph-doctor.json · skillspector-morph-init.json · skillspector-morph-module.json · skillspector-morph-share.json · skillspector-morph-update.json	2026-10-20 17:14 UTC

Known limits

What these results do not cover, kept in tests/known-gaps.md:

The guard hook is a safety net for an agent that makes mistakes. It is not a sandbox, and it does not stop an agent or script that is determined to get around it.
The guard reads Bash commands as the shell does, but it does not see inside a script file or Makefile target that an earlier call wrote, code that an interpreter decodes or builds at run time, or a command-line tool it has no rule for.
Secrets are recognised by file name and by common commands. A secret stored in a file with an unusual name is not seen.
Only Bash, file tools, Grep and MCP tool calls are checked. Other tools are left to Claude Code's own permission prompts.
The test suite is the kit's own: it shows the guard does what the kit says it does on the cases listed. No tool on this page tests the guard against hostile input beyond those cases.
No tool on this page measures quality or cost compared with plain Claude Code. Morphness adds process, and process costs tokens.
The real protection for shared environments is the host: branch protection, and environment secrets that only the owner holds.
Static scanners read the kit's text and scripts. They do not run an agent, so they say nothing about how a model behaves with these rules.

Run it yourself

git clone https://github.com/antarsoft/morphness.git && cd morphness && git fetch --tags
scripts/kit-selftest.sh          # the kit's own checks
scripts/trust-report.sh --out /tmp/trust   # everything on this page, written to /tmp/trust

Raw data for this run: results.json. Results change when the kit or the tools do, so check the date above.