Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/mbeacom/adrkit/checkgit clone --depth 1 https://github.com/mbeacom/adrkitWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00014 | $0.00568 |
| Opus 5 | $0.00007 | $0.00284 |
| Sonnet 5 | $0.00003 | $0.00114 |
| Haiku 4.5 | $0.00001 | $0.00057 |
Grade A, and why
check scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 52 lines — stays where its author put it; the contents beside it link to each section on GitHub.
adrkit — check this plan
Runs after a plan exists, and answers one question: does this plan sit inside the decisions already made, or does it quietly cross one?
This is the command the optional after_plan hook offers to run.
User Input
$ARGUMENTS
Steps
- Treat any arguments as the repo-relative paths this plan intends to touch. Pass them — the check is only as sharp as the paths it is given. With no arguments it falls back to the plan artifact itself.
- Run
{SCRIPT}with those paths. - Read the output. It is one or two marker-introduced sections, each followed
by a single JSON document:
==> adrkit:check— theCheckOutcome: which decisions govern these paths, and findings for any changed ADR. Always present.==> adrkit:evaluate— the deterministic evaluator's routing report for this plan. Present only whenADRKIT_SNAPSHOTpoints at an offline snapshot bundle. When it is absent, the script says so on stderr; that is an omission you have been told about, not a check that passed.
- Report back, in the plan's own terms:
- Every governing decision the plan touches, named by id.
- Any place the plan departs from one. Say so explicitly. A departure that is deliberate and recorded is fine; a departure nobody noticed is the thing this catches.
- If the evaluator ran, its routing outcome and any escalation trigger. The evaluator routes; it does not approve. Do not read a clean route as sign-off.
- If the plan does cross a decision, the next step is
/speckit.adrkit.draft— record the new decision rather than letting the plan silently overrule the old one.
Notes
- Read-only, and that is load-bearing rather than incidental: this command is reachable from a lifecycle hook, so it must never write.
- Exit code 1 means findings exist (corpus errors, or the evaluator returned the proposal). The report is still complete. Exit code 2 is a usage error.
- Configuration:
ADRKIT_DIR(corpus, defaultdocs/adr),ADRKIT_SNAPSHOT(evaluator snapshot bundle),ADRKIT_AS_OF(evaluation date, default today UTC),ADRKIT_CLI(explicit path to theadrentry point).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 52 lines · 14 tokens per session scan A 5c307c4c9f05
check is a command published in the GitHub repository mbeacom/adrkit (11 stars, last pushed 2d ago), licensed Apache-2.0. It adds 14 tokens to every session and 568 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
release-adr-kit
You are running the adr-kit release for version $ARGUMENTS (if empty, ask which version to release).
adr-export
Export ADRs to HTML, JSON, or PDF format.
adr-search
Search ADRs by content, status, tags, or date.
adr-setup
Interactive setup for ADR configuration in this project.
adr-supersede
Create a new ADR that supersedes an existing one.
adr-update
Update an existing ADR (status, content, or metadata).