agentleFS
Ju571nKJu571nK/lodeline/.claude/skills/verify/SKILL.md

verify

Skill17 starsChanged yesterdayApache-2.0View on GitHub

Drive the real Lodeline CLI (lodeline, lodeline-hook, lodeline-sign, lodeline-server) as short-lived processes in an isolated temp HOME and capture evidence. Use to prove a change to scan, doctor, status, hook install or enforce, or policy signing on the real binaries, not just in tests, or to reproduce a CLI bug.

189 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use instinct-export, instinct-import and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • TheKREATORSSkill

    instinct-export

    Export instincts from project/global scope to a file

    Also in 3 projects that use this

  • TheKREATORSSkill

    instinct-import

    Import instincts from file or URL into project/global scope

    Also in 3 projects that use this

  • AviadPSkill

    risk-assess

    Assess the risk and blast radius of a proposed code change before implementing it. Evaluates CI impact, framework-wide vs localized scope, rollback path, and identifies blindspots. Use when user says "what's the risk", "is this safe", "assess the impact", "risk assess", "blast radius", or before applying framework-wide or CI-affecting changes. Do NOT use for code quality review (use code-quality) or plan validation (use design-validator).

    Also in 3 projects that use this

  • diego-tobalinaSkill

    build-fix

    Fix build and compilation errors systematically. Run when build fails.

    Also in 6 projects that use this

  • AaravChadhaSkill

    secure

    Security review with a confidence gate - a finding exists only with a concrete exploit scenario and a high/medium/low confidence rating; everything else goes to a worth-hardening list, never inflated. Sweeps five surfaces: auth gates, secrets hygiene, injection and unsafe sinks, unsafe deserialization/crypto/transport, and LLM tool-use trust boundaries. Reports only, never fixes. Use when the user asks for a security review or to check vulnerabilities, secrets, or auth.

    Also in 3 projects that use this

  • ArvorCoSkill

    specify

    Create feature specification from natural language description. Use when starting a new feature or creating a spec. Triggers on: create spec, specify feature, new feature spec.

    Also in 10 projects that use this

  • +1 more from TheKREATORS

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 1 is distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 167 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Read features/README.md, then the matching feature file.

    line 10
  2. Very specific:

    Run every binary through drive.sh.

    line 14
  3. Very specific:

    Never pass a real path to --policy, --state-db, --events-dir, --control-socket, or --keystore.

    line 15
  4. Very specific:

    Never run lodeline run. See What cannot be isolated.

    line 16
  5. Very specific:

    Never run lodeline-hook install --agent antigravity --force --write.

    line 18
  6. Distinctive:

    Commit no account IDs, keys, credentials, or personal paths.

    line 21
  7. Very specific:

    Check side effects with find on the scratch roots.

    line 65
  8. Very specific:

    Keep the map current with /maintain-verification-skill.

    line 89

“Common” is measured: kept by at least 6 in 10 of the 167 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Verify Lodeline

Primary surface: the CLI. There is no server to keep alive. Build once per run, then run each drive as its own process through drive.sh, which replaces the whole environment and points every path the binaries resolve at a per-run scratch directory.

Read features/README.md, then the matching feature file. A proof covers every entry point the feature file lists, or names the ones it skipped.

Hard rules

  • Run every binary through drive.sh. Never run target/debug/lodeline* directly. A direct run reads the real ~/.claude, ~/.codex, and ~/.cursor.
  • Never pass a real path to --policy, --state-db, --events-dir, --control-socket, or --keystore. Point them under ../ of the drive's working directory. doctor --state-db opens the file it is given.
  • Never run lodeline run. See What cannot be isolated.
  • lodeline-hook install and uninstall stay preview, without --write, unless drive.sh runs them. Inside drive.sh, --write lands in the temp HOME.
  • Never run lodeline-hook install --agent antigravity --force --write. It calls the external agy CLI, whose state this harness does not isolate.
  • Lodeline measures and does not block by default. Blocking journeys are opt-in, use --enforce, and run only through drive.sh. A default-path proof must show that nothing was blocked.
  • Kill only what the run started. drive.sh cleanup does that by recorded pid and start time. Never use pkill, killall, or a kill by process name.
  • Commit no account IDs, keys, credentials, or personal paths. Evidence holds real home paths (see below), so quote it in a PR only after you replace them. AWS test-node procedures stay out of this skill.

Launch

From the repo root or a worktree root:

RUN=$(.claude/skills/verify/drive.sh new)

This runs cargo build --locked for lodeline-agent, lodeline-hook, lodeline-signer, lodeline-server, and lodeline-sender, creates the empty run, and prints the run id on stdout. The instance is ready when the id prints and Doctor passes. No migration or seed step exists: scratch starts empty, and each feature file seeds what it needs.

The shell does not keep RUN between tool calls, so pass the id literally in each later call. Every run has its own scratch directory, so runs can overlap, except for the fleet port in features/fleet-local.md.

Doctor

.claude/skills/verify/drive.sh doctor "$RUN"

Read-only. It exits 0 when each binary reports the workspace version, no source changed since the build, lodeline resolves state.db inside scratch, and each background process of this run is still up. After a [FAIL] about changed source, start a new run. Run Doctor first whenever anything looks off.

Drive

.claude/skills/verify/drive.sh run "$RUN" <label> -- <command> [args...]
.claude/skills/verify/drive.sh bg "$RUN" <label> -- <command> [args...]
  • run executes one command in scratch/project and returns its exit code. It prints the first 40 lines of stdout and the first 20 of stderr. Stdin passes through.
  • bg starts a long-lived process, for lodeline-server only, and records its pid.
  • Inside a drive, $HOME, $XDG_STATE_HOME, $XDG_CONFIG_HOME, $XDG_RUNTIME_DIR, and $TMPDIR are scratch paths, and PATH is target/debug, /usr/bin, and /bin. Use sh -c '...' with single quotes to seed fixtures, so those variables expand inside the sandbox.
  • A label is single-use, so evidence is never overwritten.

drive.sh refuses root, because as root the binaries use /var/run, /var/lib, and /etc with no override. It is macOS and Linux only. The Windows paths have no driver.

Evidence

Each drive writes <label>.cmd, <label>.stdout, <label>.stderr, and <label>.exit to ${TMPDIR:-/tmp}/lodeline-verify/<run>/evidence/, outside the repo. build.txt there records the git SHA, the dirty files, and the binary versions. drive.sh refuses a TMPDIR inside the repo. Evidence survives cleanup and stays until the OS clears TMPDIR, so copy it elsewhere if it must last.

Proof standards:

  • Drive the real user path: the shipped binary and its flags. Test seams such as report_json_for are not proof.
  • Capture the action and the resulting state. Pair a command with a read-only second view, such as lodeline-hook verify after install --write.
  • Check side effects with find on the scratch roots. A preview proof shows that nothing was written.
  • Mock only at a production boundary. Fixture configs under the temp HOME are fine. A stubbed verdict is not proof of a daemon decision.
  • Report each result as verified, skipped, or failed. A skipped entry point is not verified through another path.

Cleanup

.claude/skills/verify/drive.sh cleanup "$RUN"

Stops this run's background processes, removes scratch/, keeps evidence/, and prints what survived. It is safe to run twice. Run it after every failed attempt too.

What cannot be isolated

  • User enumeration. lodeline doctor and lodeline show paths take users from /Users on macOS and /etc/passwd on Linux, not from $HOME. doctor checks that real users' config paths exist and prints those paths, so its evidence holds real home paths. It does not read their contents.
  • System policy files. lodeline scan reads managed settings at /Library/Application Support/ClaudeCode on macOS, /etc/claude-code on Linux, and /etc/codex, read-only. On a host with those files, extra rows or controls appear. Assert on your fixture's row, not on the whole table.
  • Full Disk Access. doctor on macOS probes the system TCC database path. The result, and so the exit code, depends on the host.
  • The daemon. lodeline run watches every enumerated user's real configs. A policy can disable a target by id, but no run has shown that this covers every daemon read. Daemon behavior stays with cargo test -p lodeline-agent. Anything that needs a live daemon is out of reach here: show stats, show risk, show policy-status, show targets, reload, live AI Guard data in doctor, the agent: running row of status, and real deny verdicts.

Secondary surfaces

  • Fleet stack. lodeline-sign and lodeline-server in plain HTTP on loopback run locally, as features/fleet-local.md shows. Enrolling a host over mTLS with lodeline-sender enroll runs locally too, as features/fleet-enroll.md shows. Enrollment on a packaged host (owners, systemd, SELinux) has no recipe here. The lodeline-manager console has its own verify skill in apps/manager/.claude/skills/verify/.
  • Diagnostic app. The macOS Tauri app is planned and has no driver yet.

Keep the map current with /maintain-verification-skill.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.