Borrowing it
Nothing to install: this file belongs to shaahink/conductor. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/shaahink/conductor/master/.claude/skills/watch-run/SKILL.mdgit clone --depth 1 https://github.com/shaahink/conductorWrote 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/skills/shaahink/conductor/watch-run)<a href="https://agentmods.dev/skills/shaahink/conductor/watch-run"><img src="https://agentmods.dev/badge/skills/shaahink/conductor/watch-run/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/shaahink/conductor/watch-run"><img src="https://agentmods.dev/badge/skills/shaahink/conductor/watch-run.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00051 | $0.00974 |
| Opus 5.5 | $0.00020 | $0.00390 |
| Sonnet 5.5 | $0.00010 | $0.00195 |
| Haiku 4.5 | $0.00005 | $0.00097 |
Grade A, and why
watch-run 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 2mo 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 — 63 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the night watch for a live conductor run. Your job is cheap vigilance: sit silent
between events, wake only on the filtered signals below, intervene with control verbs only when a
rule fires, and keep every reaction to 1–3 lines. You are NOT the delivering agent — never edit
the repo, never commit, never start/stop sessions by hand beyond the verbs listed here.
Arm the watch (one Monitor, persistent)
The engine appends every meaningful line to <repo>/.conductor/conductor.log regardless of how the
run was started (owner terminal with Face, or headless). Default repo: C:/code/conductor-baton,
default plan: plans/conductor-ux.plan.json — confirm with the user only if the prompt says
otherwise.
Arm exactly one persistent Monitor:
tail -F "C:/code/conductor-baton/.conductor/conductor.log" | grep -E --line-buffered "session #[0-9]+ (start|exited|rolled over)|verifier (score|passed|failed|produced)|NEEDS HUMAN|needs-attention|parked|stall|backing off|usage limit|abort|cancelled|state saved|WARNING|circuit|--max-sessions|crash|discarded"
Then tell the user the watch is armed and go quiet. Do not poll. Do not schedule wakeups.
Intervention rules (check top to bottom on each event batch)
The plan's own rails already bound a runaway: per-stage attempts cap at 2× the stage's session
budget then park; maxResumesPerSession 3; sessionTimeoutMinutes 90; maxSessions 40 then stop;
repeated agent-backend refusals park NeedsHuman after MaxBackoffs. Your job is the early cut and
the phone call, not re-implementing those rails.
- Quota guard — any
usage limit/backing offline: first one, note it silently. Second within the same run:conductor pause -p <plan>and send a PushNotification ("run paused — agent backend refusing, check quota"). This is the drain-a-Claude-plan-in-a-loop scenario. - Churn loop — the same stage shows
Fix/Verifyattempt lines reachingattempt N/Mwith N ≥ 4, or 3 consecutive sessions exit in under ~2 minutes each:conductor pause -p <plan>, PushNotification ("run paused — is churning attempts without progress"). - Park —
NEEDS HUMAN/parked/AwaitingOwner: the engine already stopped spawning. PushNotification with the reason line. No verb needed. - Run ended —
state saved/--max-sessions/abort: PushNotification with a one-line summary (conductor status -p <plan>for checkpoint counts — it is offline and cheap by default). - Healthy traffic (session start/exit, gates PASS, verifier passed, workflow steps): no reply at all, or at most one line if several checkpoints just confirmed. Silence is the correct output for a healthy run.
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.
- 2mo ago First seen · 63 lines · 51 tokens per session scan A 65538dce559a
watch-run is a skill published in the GitHub repository shaahink/conductor (5 stars, last pushed 1mo ago), licensed MIT. It adds 51 tokens to every session and 974 once invoked, about $0.0002 per session on Opus 5.5. 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 skills, from other repositories
cratis-chronicle-client-dotnet
Use when connecting a standalone .NET console, worker or service directly to Chronicle or diagnosing its client lifecycle. Not for Arc integration, kernel implementation or another language's client.
cratis-fundamentals-concept
Use when a C# or TypeScript domain value needs a distinct type, a concept generates a value or an identity addresses a Chronicle stream. Not for enums, DTO-only values, Go concepts or event migration.
analyzing-dotnet-performance
Use when reviewing .NET hot paths or scanning source for allocation-heavy and inefficient patterns. Not for measuring a running process, benchmark design or SIMD optimization.
csharp-refactoring
Use when restructuring C# code without changing behavior or separating a requested refactor from behavioral changes. Not for ordinary features, bug fixes, dependency upgrades, test additions or formatting alone.
directory-build-organization
Use when projects repeat MSBuild policy or an existing Directory.Build hierarchy has import, placement or layering defects. Not for isolated property semantics, legacy migration or a lone project without shared build files.
extension-points
Use when MSBuild imports or hooks are missing, ordered incorrectly or overwritten, including NuGet build assets. Not for a single Import guard, target incrementality or props-versus-targets placement.