probe-run
Skill0 starsChanged 2 months agoApache-2.0View on GitHub
Run one or more dynamic-verification probes against a local ephemeral docker-compose mirror of the target system. Captures observation traces as committed artefacts at `lineage/{repo}/probe-runs/{date}-{probe_id}.yaml` and merges measured values back into `feature-flows.yaml`, `test-map.yaml`, and per-sidecar `confidence_per_field` annotations. Local-only execution per APPROACH.md section 5 rule 12 + section 9 — no remote infrastructure ever. Slice 2 of `adrs/drafts/dynamic-verification-layer.md`.
In this repository27 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- alignSkill
- code-walkSkill
- conceptsSkill
- contributeSkill
- coverageSkill
- doc-gap-checkSkill
- enrichSkill
- enumerateSkill
- find-implicit-adrsSkill
- implementSkill
- ingest-docsSkill
- log-issueSkill
- navigateSkill
- next-batchSkill
- orientSkill
- panelSkill
- probe-defineSkill
- probeSkill
- reflect-featureSkill
- retrieveSkill
- reviewSkill
- scanSkill
- statusSkill
- streamsSkill
- test-coverageSkill
- triageSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 2 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
If any prerequisite fails, the skill reports the missing piece and exits.
- Very specific: line 51
Read lineage/{repo}/manifest.yaml → confirm last_scan_commit.
- Very specific: line 52
Read the probe definition(s) at lineage/{repo}/probes/{probe-id}.yaml.
- Very specific: line 53
Check verified_against_commit on each probe vs last_scan_commit.
- Very specific: line 54
Verify Docker daemon is reachable: docker ps (zero-exit).
- Very specific: line 55
Verify the runner is installed: ls lineage/_extractor/probe-runtime/runner.py (exists, executable).
- Very specific: line 59
Run python3 lineage/_extractor/probe-runtime/runner.py <probe-id> --validate for each selected probe.
- Distinctive: line 69
After each PASS/FAIL run the runner performs two automatic merges:
- Very specific: line 80
Run docker ps --filter "name=probe-" to confirm no leftover probe containers.
- Very specific: line 96
Cleanup is mandatory: Every probe-run terminates with the stack torn down (docker-compose down -v).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Run a dynamic-verification probe (slice 2)
Drive the layer-5 dynamic-verification pipeline. Each invocation executes one (or several) declarative probe definitions against an ephemeral local docker-compose stack, captures measurements, and feeds them back into the static-layer artefacts.
This skill is the maintainer-facing entry point per adrs/drafts/dynamic-verification-layer.md slice 2.
Prerequisite
- Local Docker daemon running and reachable as the current user (
docker pssucceeds without sudo). python3withPyYAML+requestsinstalled.- The probe-stack docker-compose file exists at
lineage/_extractor/probe-stacks/{profile}.docker-compose.yml. - The probe definition exists at
lineage/{repo}/probes/{probe-id}.yamland is well-formed (runner.py --validate <id>passes). - The substrate has been scanned at least once (
lineage/{repo}/manifest.yamlexists withlast_scan_commit).
If any prerequisite fails, the skill reports the missing piece and exits.
Argument forms
| Form | Behaviour |
/probe-run <probe-id> | Execute one probe; produce one probe-run artefact; merge measured values into the static layers. |
/probe-run <id-1> <id-2> ... | Execute multiple probes in batch mode (shared docker-compose lifecycle). All listed probes must share the same stack_profile. Slice-4 capability. |
/probe-run --feature <feature-id> | Slice-5. Resolve feature_id against lineage/{repo}/probes/*.yaml; run every matching probe in batch mode. Mutually exclusive with positional probe IDs. |
/probe-run --batch <id-1> ... | Explicit batch-mode flag. Implied when multiple positional IDs are given OR when --feature matches more than one probe. |
/probe-run --dry-run <probe-id> | Validate + parse + report what WOULD execute, but do not bring up the stack. |
/probe-run --show <probe-id> | Slice-6. Read-only. Print the probe definition + the most recent probe-run artefact's outcome, observed values, and assertion-pass count. No execution, no merge. |
/probe-run --validate <probe-id> | Read-only. Parse probe + scope-check; exit 0 if well-formed, non-zero otherwise. |
/probe-run <probe-id> --no-merge | Execute the probe but do NOT merge measured values back into the static layers. Useful for iterating on probe design. |
/probe-run <probe-id> --no-summary | Slice-5. Skip the per-batch summary artefact + investigator-log append. Useful when running probes in tight iterative loops. |
/probe-run <probe-id> --allow-stale | Slice-6. Override the substrate-staleness gate (ADR Rule 5) — execute even if verified_against_commit lags the substrate's last_scan_commit by more than 5 commits. The runner refuses by default with FATAL when lag > threshold; pass this flag if the maintainer has independently verified the probe still applies. |
Slice-5 batch outputs
In batch mode (multiple probes, --batch, or --feature), the runner emits two additional artefacts on top of the per-probe probe-runs/{date}-P-NNN.yaml files:
- Per-batch summary at
lineage/{repo}/probe-runs/{date}-batch-{trigger-slug}.md— markdown table of per-run outcomes + per-feature aggregation (which test-classes per feature this batch empirically covered). - Investigator-log append — a new
## Probe-runs YYYY-MM-DD — <trigger>section appended tolineage/{repo}/investigator-log.md, cross-referencing the per-batch summary. Per dynamic-verification ADR slice 5 ("each batch's investigator-log entry now carries a probe-runs section alongside reducer diffs"). Idempotent on (date, trigger) — re-running replaces the existing section in place.
Protocol
1. Orient (read-only — no execution yet)
- Read
lineage/{repo}/manifest.yaml→ confirmlast_scan_commit. - Read the probe definition(s) at
lineage/{repo}/probes/{probe-id}.yaml. - Check
verified_against_commiton each probe vslast_scan_commit. If lag > 5 commits and--allow-staleis not set, abort withSTALE_PROBEand surface the offset. - Verify Docker daemon is reachable:
docker ps(zero-exit). - Verify the runner is installed:
ls lineage/_extractor/probe-runtime/runner.py(exists, executable).
2. Validate (per probe — also a no-execution gate)
Run python3 lineage/_extractor/probe-runtime/runner.py <probe-id> --validate for each selected probe. If validation fails on any probe, stop. Report which probes failed; do not proceed.
3. Execute (per probe — serial unless future --parallel)
Spawn the probe-runner subagent via the Agent tool with the probe ID(s) + maintainer options. The subagent does the actual python3 runner.py <probe-id> invocation, captures the runner's exit code + stdout, and emits the probe-run artefact.
For --show and --validate forms, no subagent spawn is needed — the skill directly invokes the runner with the appropriate flag.
4. Merge (unless --no-merge)
After each PASS/FAIL run the runner performs two automatic merges:
- Sidecar confidence merge (slice 4) — appends a
## probe_verificationsentry to every contributing sidecar of the probe'sfeature_id(per dynamic-verification ADR Rule 4 — closes the layer-5 → layer-2 feedback loop). Idempotent onprobe_run_id. - Feature-flows stamp (slice 6) — appends a
probe_verifications:list entry inside the matching feature's block inlineage/{repo}/feature-flows.yaml. Replace-by-probe-id semantic: feature-flows.yaml carries the CURRENT measured state per probe (latest run); the full run history lives inprobe-runs/. The narrativeobserved_vs_expected.facetsblock stays maintainer-authored; the auto-stamp is the audit trail (run-IDs, outcomes, test-classes empirically covered, commit pinned).
Still maintainer-authored (slice-7+ candidates):
lineage/{repo}/test-map.yaml— per-feature matrix cell flips toPROBED-PASSING/PROBED-PINNING-BUG/PROBE-TEST-DISAGREEMENT.
5. Cleanup verification
Run docker ps --filter "name=probe-" to confirm no leftover probe containers. If any remain, surface the manual cleanup command and ask the maintainer to confirm before re-running.
6. Report
Concise output:
- Counts:
<N> probes executed; <P> PASS, <F> FAIL, <E> ERROR, <T> TIMEOUT, <S> SCOPE_VIOLATION - Probe-run artefact paths
- Static-layer merges (which feature_id, which test_class cells flipped, how many sidecar confidence annotations)
- Top-1 headline finding (e.g. "P-002 PASS — anonymous detail-read under DISABLED inflates view_count by 10/10 calls — REFACTOR-073 confirmed at the empirical level")
- Cleanup status
Rules
- Local-only execution. Per APPROACH.md section 5 rule 12 + dynamic-verification ADR Rule 1. No remote URLs in probe definitions; no remote services in the docker-compose stack profiles; nothing outside the maintainer's local docker daemon. The runner enforces ALLOWED_VERBS; the subagent enforces no-arbitrary-Bash; this skill enforces the prerequisite checks.
- Probe versioning aligned to substrate.
verified_against_commiton every probe MUST equal (or lag by ≤5 commits) the substrate'slast_scan_commit. Stale probes are refused unless--allow-staleis explicit. - Cleanup is mandatory. Every probe-run terminates with the stack torn down (
docker-compose down -v). If a run aborts before cleanup, the subagent emits a cleanup command for the maintainer to run. - No source-repo modification. This skill (via the runner + subagent) never modifies files outside
lineage/{repo}/probe-runs/,lineage/{repo}/feature-flows.yaml,lineage/{repo}/test-map.yaml, and per-sidecar confidence_per_field annotations underlineage/{repo}/understanding/. Substrate files (manifest.yaml,nodes.jsonl,edges.jsonl) and target-repo source files are never touched.
Cross-references
- ADR anchor:
adrs/drafts/dynamic-verification-layer.md - Subagent:
.claude/agents/probe-runner.md - Runner:
lineage/_extractor/probe-runtime/runner.py - Stack profiles:
lineage/_extractor/probe-stacks/ - First-experiment probes (slice 2):
lineage/odd-platform/probes/P-001.yaml,P-002.yaml,P-003.yaml - Trigger case-law:
retrospectives/LSN-017-per-node-scan-cannot-see-cross-layer-user-effects.md
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.

