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/akahkhanna/groundtruth/groundtruth-blockgit clone --depth 1 https://github.com/akahkhanna/groundtruthWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/akahkhanna/groundtruth/groundtruth-block)<a href="https://agentmods.dev/commands/akahkhanna/groundtruth/groundtruth-block"><img src="https://agentmods.dev/badge/commands/akahkhanna/groundtruth/groundtruth-block.svg" alt="Measured on agentmods" height="20"></a>What 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 | $0.00018 | $0.01114 |
| Opus 5 | $0.00009 | $0.00557 |
| Sonnet 5 | $0.00004 | $0.00223 |
| Haiku 4.5 | $0.00002 | $0.00111 |
Grade A, and why
groundtruth-block 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 5d ago.
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 — 31 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Set Groundtruth's block mode based on the argument: $ARGUMENTS (on or off; empty ⇒ just report the current value).
off or empty
- off → read
.claude/groundtruth/config.json(preserve any other keys), write"block": false, confirm. - empty → report the current
blockvalue. Note if envGROUNDTRUTH_BLOCKis set — it overrides config (the un-disablable anchor).
on — never flip silently; confirm what will START HALTING first
The rule is: block must never enforce a predicate the user hasn't seen. So before writing anything, show an itemized, evidence-backed picture, then require an explicit yes.
-
Read
.claude/groundtruth/compiled-rules.json(the armed rules) and, if present,.claude/groundtruth/history.jsonl(the append-only per-turn log — each finding carriescls,sev, and, for a compiled rule,rule= its id). -
Compute per-rule fire counts from history: tally findings by their
ruleid across all lines (older lines predating id-logging won't have it — count what's there and say the window is "since the log carried ids"). This is the evidence — reviewing behavior ("fired 14×, unarmed twice" vs "fired 0× in 6 weeks") is tractable where re-reading regexes is not. Caveat the counts as advisory:history.jsonlis gitignored and agent-writable (not integrity-snapshotted), so a count can be forged — it can't hide a rule (the list below comes fromcompiled-rules.json, which is authoritative), only misstate a number. So never let a low count alone justify enabling a block rule; the count informs, the itemized list decides. -
Show the user what block actually changes — be honest about the severity gate (block only enforces findings that are already
block-severity; it does NOT auto-promote your warn rules):- Built-in block checks that will now HALT (always, once block is on): hardcoded secret, committed
.env, RLS-off / anon-permissive policy, and a contractCA(claimed-but-absent) — a claim the diff/transcript don't support: atests_pass/build_passthat never ran or ran red, acreated/modified/deletedfile absent from the diff, or a claimed symbol not defined in the added code. These are code, not your rules — always block-severity. Plus anNC(no valid manifest on a code turn) when the repo is contract-aware —/groundtruth-setuphas written thegroundtruth-claimsinstruction intoCLAUDE.md; awareness is baseline-anchored (content and file tree), so a mid-turn strip or untrack of the instruction can't downgrade it. (UCstays warn.) - Your armed rules at
blockseverity — the rules that will start refusing commits. List each:id· message · fired N× · last-fired. Flag any with 0 fires or that the user hasn't visibly reviewed: "never fired / unreviewed — look before enabling." - Your armed rules at
warnseverity — state plainly: these stay warn and will NOT halt under block mode. Optionally list the top few by fire count as promotion candidates (make one halt withseverity:"block").
- Built-in block checks that will now HALT (always, once block is on): hardcoded secret, committed
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.
- 5d ago First seen · 31 lines · 18 tokens per session scan A 7894cd6502cf
groundtruth-block is a command published in the GitHub repository akahkhanna/groundtruth (6 stars, last pushed 1mo ago), licensed MIT. It adds 18 tokens to every session and 1,114 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 2026-08-31.
Other commands, from other repositories
checklist
Generate a custom checklist for the current feature based on user requirements.
clarify
Identify underspecified areas in the current feature spec by asking up to 5 highly targeted clarification questions and encoding answers back into the spec.
specify
Create or update the feature specification from a natural language feature description.
analyze
Perform a non-destructive cross-artifact consistency and quality analysis across spec.md, plan.md, and tasks.md after task generation.
converge
Assess the current codebase against the feature's spec, plan, and tasks, then append any remaining unbuilt work as new tasks to tasks.md so implement can complete it.
implement
Execute the implementation plan by processing and executing all tasks defined in tasks.md.