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/mohammad19974/bounded-relay/speckit.boundedrelay.execute-wavegit clone --depth 1 https://github.com/mohammad19974/bounded-relayWrote 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/mohammad19974/bounded-relay/speckit.boundedrelay.execute-wave)<a href="https://agentmods.dev/commands/mohammad19974/bounded-relay/speckit.boundedrelay.execute-wave"><img src="https://agentmods.dev/badge/commands/mohammad19974/bounded-relay/speckit.boundedrelay.execute-wave.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 | $0.00013 | $0.00721 |
| Opus 5 | $0.00006 | $0.00360 |
| Sonnet 5 | $0.00003 | $0.00144 |
| Haiku 4.5 | $0.00001 | $0.00072 |
Grade A, and why
speckit.boundedrelay.execute-wave 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 4d 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.
What it actually says
Execute One Routed Wave
Treat execution.json and the verified routing.json named by $ARGUMENTS as
the complete authority envelope. Work on the single activeWave; never batch a
future wave or broaden a path lease.
- Verify both evidence files, the active baseline, repository rules, and every dependency result. If the active wave already contains a complete result for every routed task, make no provider call and no code change. This makes a paused loop safe to resume.
- Process read-only tasks first in canonical task-id order. A Claude-host task
is handled directly with the host-selected Claude model. A Codex task must go
through BoundedRelay, use its allowlisted model policy, poll with
afterRevision, and retrieve the terminal result once. - Execute the wave's single writer, if present, only after all read-only work
in the wave is accepted. For Claude-host, write only inside its lease. For
Codex, call
codex_worker_proposeagainst the exact active baseline, inspect the patch and changed paths, then integrate it as the sole host writer. Never ask BoundedRelay to apply, commit, merge, push, publish, or deploy. Persist the exact returned patch bytes atpatches/<taskId>.patchbeneath the run directory with owner-only permissions; record that relative file and its recomputed digest. The patch is local run evidence and must never be added to Git. - Run focused repository checks against the completed working-tree content.
Record 1-64 redacted check receipts: safe profile and label, SHA-256 of the
exact argv, relative cwd, zero exit code, output digests, the exact tested
Git tree ID, and timestamps. The tested tree must equal the following
authorized checkpoint commit tree. Never persist raw output, environment
values, tokens, or secrets. With a project profile, every required check ID
must have a successful receipt whose
profile,cwd, andcommandSha256exactly match the sealed definition. Extra receipts are allowed only for check IDs defined by that profile. Across all writer results, required and optional receipts together must not exceed 256; refuse the active wave before checkpointing if it would add receipt 257. The coordinator may run reviewed commands; no BoundedRelay workflow script executes profileargvautomatically. - Append exactly one typed result per active task. Preserve every engine-owned
field. Results include routed provider/wave, effect, baseline, provenance,
the worker-observed model and reasoning effort, verification, timestamps, and
empty checks for analysis. Codex model fields must exactly match the routed
policy; Claude fields remain null because the host selection is not
independently observable. A write result also includes exact changed files
and checks; a Codex write includes job and patch digests and uses
proposal-integratedonly after host inspection and integration. - Stop with the worktree uncommitted when code changed. The following human gate owns inspection and checkpoint authorization. Do not advance the wave, alter routing, repair unrelated code, or manufacture missing evidence.
$ARGUMENTS
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.
- 4d ago First seen · 57 lines · 13 tokens per session scan A 282c73a85ee3
speckit.boundedrelay.execute-wave is a command published in the GitHub repository mohammad19974/bounded-relay (1 stars, last pushed 6d ago), licensed MIT. It adds 13 tokens to every session and 721 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
tokendiet
Audit your code for wasteful LLM API spend and get a concrete plan to cut it — cache, batching, retry caps, prompt bloat. No fabricated numbers.
config-edit
Edit llm-redact config with the dashboard editor's guardrails: change rules, modes, deny strings, allowlists, NER, providers; validate; hot-reload.
users
List llm-redact named users and seats; guide invites and revokes.
audit
Verify the llm-redact tamper-evident audit chain.
config-show
Show the effective llm-redact configuration (env overrides named).
doctor
Run llm-redact's read-only diagnostics (config, build, vault, extras, posture).