test-skills
Skill3 starsChanged 10 days agoMITView on GitHub
runs every skill's sidecar at a shallow depth and proves each still answers (saves report to .construct/)
In this repository3 other files it gives its agents.
- export-readmeSkill
- push-releaseSkill
- validate-skillsSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 6 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 23
no argument runs every sidecar under plugins/*/skills/ and .claude/skills/.
- Very specific: line 31
IF --quick, STOP after the inline report.
- Distinctive: line 33
READ the table and group the rows by tier, since one broken shared file fails many skills at once
- Very specific: line 35
TREAT a t0 row as the most urgent, since a file that will not parse never ran at all
- Very specific: line 36
TREAT a t3 row as a lie the exit code told, since the case returned 0 and proved nothing
- Distinctive: line 38
RUN the handover's per-skill line to read one failure in full, rather than re-running the tree
- Distinctive: line 39
STOP after reporting; a repair is the user's to approve, and the tool mutates nothing itself
- Very specific: line 48
CREATE the file first if it does not exist, with # <audit_file> as its only line
- Very specific: line 122
RUN .claude/skills/test-skills/test-skills.sh after moving a shared file or renaming any path
- Very specific: line 145
every field prints, in this order.
- Distinctive: line 146
every value is COPIED from the source named beside it, never composed fresh
- Distinctive: line 147
ask what they are actually trying to do, and what they have already tried
- Distinctive: line 149
run no step, write no file, and never fall through to step 1
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Instructions
Telemetry
.claude/skills/test-skills/test-skills.sh "$ARGUMENTS"
echo "sidecar exit: $?"- it already ran, so there is no command to issue
- no argument runs every sidecar under
plugins/*/skills/and.claude/skills/; a path scopes it scanned: N sidecar(s)→ what the run covered, which is one sidecar per skill directoryadopted: N without a --test case→ skills still outside the contract, and each one is a WARNhelp: requestedfrom a--helpinvocation → read## Helpand run nothing- fail (
sidecar exit> 0) → a tier failed, or a WARN landed under--strict; quote the findings table verbatim inside a markdown code block, and fix nothing until the user has read it - success (
sidecar exit= 0) → reporterrorsandwarnings, then work the table top to bottom --quickreports inline and writes nothing,--strictpromotes warnings to errors- IF
--quick, STOP after the inline report;audit_file: noneconfirms it, and nothing is appended
- READ the table and group the rows by tier, since one broken shared file fails many skills at once
- NAME the tier each failure landed in, since the tier says how far the sidecar got before it broke
- TREAT a
t0row as the most urgent, since a file that will not parse never ran at all - TREAT a
t3row as a lie the exit code told, since the case returned 0 and proved nothing - QUOTE the
detailcolumn verbatim for every FAIL, since it carries the path or the repair - RUN the handover's per-skill line to read one failure in full, rather than re-running the tree
- STOP after reporting; a repair is the user's to approve, and the tool mutates nothing itself
the artifact
append one entry to [audit_file], in the shape under ## the entry below:
- the heading reads
## Smoke Audit #[next_audit]: [timestamp], both taken from the telemetry stateis the scan counts and the probes directory, as hyphen bullets, one clause eachresultsis the skill/tier/result/detail table the sidecar printed, as a markdown tabletelemetryis the sidecar's whole output, fenced and unedited, pasted last- CREATE the file first if it does not exist, with
# <audit_file>as its only line - the sidecar names the path and the count and writes nothing, so a ci run leaves no entry
- the run writes nothing under that directory itself, so an absent file means nobody recorded a run
the entry
the artifact this skill appends to;
/gitgud:auditgrades the shape on its next run
.construct/maintainer/test-skills/YYYY-MM-DD.md
one file per day, appended to by every run:
- an entry captures one run at a moment in time, so it is never edited after the fact
- lines are hyphen bullets holding a single clause, capped at 100 characters
Smoke Audit #1: YYYY-MM-DD HH:MM
state
the sidecars scanned, the two counts, the skills still without a case, and the probes directory
example:
- scanned: 32 sidecar(s), 0 error(s), 0 warning(s)
- probes: .construct/maintainer/test-skills/probes, kept: no
results
the table the sidecar printed, one row per tier per skill
example:
| gitgud:audit | t3-artifact | ok | 6 check(s), 0 note(s), 0s |
telemetry
the sidecar's whole output, fenced and unedited
the tiers
each tier runs only when the one before it passed, so one break reports once rather than four times:
t0parses the file:bash -n, the shebang, shellcheck at error level, and the tracked exec bitt1asks it two questions:--helpprints its marker, and an unknown flag is refused- a refusal is any non-zero exit, since
fatal:andusage:both refuse and the wording is nobody's - a confirm gate refuses too, by pricing the run and stopping before it ever parses a flag
- the flag question is asked only of a skill whose
argument-hintdeclares nothing past its flags - a skill taking prose or a path reads a dash-leading word as its own argument, and is right to
- the hint is the declaration, so the doc decides which skills the question even applies to
t2resolves what the sidecar declares about itself, reading its text and never running itt3runs--test, which is the only tier that executes the skill's file at all- nothing here does the skill's work; a real run belongs to the trigger a user invokes on purpose
- a
t0-moderow is the break that only ever lands on someone else's machine, so it is an error here
what t2 resolves
every check reads the sidecar's own text, so a broken reference is named whether or not it runs:
sourcestats every# shellcheck source=path, since a sourced file dying takes the plugin down- ci runs
shellcheck --severity=warning, where a missing source is SC1091 at info and passes - so this tier is the only gate in the repo that fails a sourced file that moved or was deleted
seestats every path on the@seeline, which is how a rename that missed one gets caught- a missing FILE on that line is a failure, and a missing DIRECTORY is a note, never a failure
- an artifact directory is written by a run nobody has made yet, so its absence proves nothing
toolresolves every binary the file guards on withcommand -v, naming a machine gap as one
the --test contract
every sidecar answers --test the same way, so the case is COPIED rather than written:
case " $* " in *" --test "*) echo "test: ok"; exit 0;; esac- it sits directly below the
--helpcase, above every preflight and every confirm gate - a test that has to pay for a confirm is a test nobody runs, which is why it sits that high
- the line is identical in all 32 sidecars: no path, no invocation name, nothing to paste wrong
- it proves the file parses and the guards above it returned, and deliberately nothing more
- the checks that need a sidecar's declarations live in
t2, which reads them from outside - one shared helper for this would be three byte-identical copies, one per plugin, to print a marker
Verify
not part of the trigger; these run after a doc or a sidecar changes
- RUN
.claude/skills/test-skills/test-skills.shafter moving a shared file or renaming any path - EXPECT zero warnings, since ci runs
--strictand every sidecar already carries the case - CONFIRM a new sidecar carries the
--testcase, sincevalidate-skillswarns and does not block - CONFIRM this sidecar reads
100755undergit ls-files -s, since the disk bit lies - FIX every
t0-moderow, since that break reaches only the people who installed from a clone - RERUN a disagreeing
t3row scoped to that one path, since this sidecar writes no scratch to keep
Help
IF the invocation carries
--helpor-h, this section is the whole turn:
SKILL: /plugin:name
DESCRIPTION: <the `description` frontmatter, verbatim>
POSTURE: <the readme index's keyword for this skill>
FLAGS:
- --flag: <what it changes, in the telemetry bullet's own words>
ARGUMENTS:
- <arg>: <what it names>
ARTIFACT: <the `metadata.artifact` path, or none>
OUTPUT: <what lands in the turn: an audit entry, a handover block, an inline report>
SPEC: <this doc's own path>- every field prints, in this order; one with nothing to say prints
none - every value is COPIED from the source named beside it, never composed fresh
- ask what they are actually trying to do, and what they have already tried
- name the flag or the sibling skill that fits their answer, then STOP
- run no step, write no file, and never fall through to step 1
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

