agentleFS
prime-radiant-incprime-radiant-inc/greenfield/skills/container-execution/SKILL.md

container-execution

Skill284 starsChanged 6 months agoApache-2.0View on GitHub

Infrastructure skill for containerized target execution. Runtime detection, container lifecycle, security restrictions, interaction patterns.

  • Deletes or force-pushes
  • Installs packages

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    All runtime analysis targets run inside containers.

    line 8
  2. Distinctive:

    If neither Docker nor Podman is available, skip all agents that require containers.

    line 23
  3. Distinctive:

    Does not leak the target's identity

    line 30
  4. Very specific:

    If the build fails, log the error to workspace/raw/runtime/build-log.txt and mark runtime mode as unavailable.

    line 121
  5. Very specific:

    All target interaction goes through docker exec (or podman exec).

    line 163
  6. Distinctive:

    Before executing commands, verify:

    line 202
  7. Very specific:

    All captured output goes to workspace/raw/runtime/

    line 279
  8. Distinctive:

    Avoid including raw credential values in captured output (use judgment, not regex)

    line 280

≈ “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.

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

Container Execution

All runtime analysis targets run inside containers. Never execute untrusted code on the host.

Runtime Detection

if command -v docker &>/dev/null; then
  RUNTIME="docker"
elif command -v podman &>/dev/null; then
  RUNTIME="podman"
else
  echo "No container runtime found. Runtime observation unavailable."
  # Continue with static analysis modes only
fi

If neither Docker nor Podman is available, skip all agents that require containers. This is not an error — runtime observation is additive.

Container Naming

Container name: greenfield-${WORKSPACE}-target

  • Deterministic (same workspace = same name)
  • Does not leak the target's identity
  • Allows multiple concurrent analyses

Container Lifecycle

digraph container_lifecycle {
    rankdir=TB;

    "Start container lifecycle" [shape=doublecircle];
    "Container runtime available?" [shape=diamond];
    "Build target image" [shape=box];
    "Build succeeded?" [shape=diamond];
    "Start container with resource limits" [shape=box];
    "Verify container is running" [shape=box];
    "Container running?" [shape=diamond];
    "Execute agent commands via docker exec" [shape=box];
    "Stop and remove container" [shape=box];
    "Lifecycle complete" [shape=doublecircle];
    "Skip runtime mode, continue static analysis" [shape=ellipse];
    "Log failure, skip runtime mode" [shape=ellipse];

    "Start container lifecycle" -> "Container runtime available?";
    "Container runtime available?" -> "Build target image" [label="yes"];
    "Container runtime available?" -> "Skip runtime mode, continue static analysis" [label="no"];
    "Build target image" -> "Build succeeded?";
    "Build succeeded?" -> "Start container with resource limits" [label="yes"];
    "Build succeeded?" -> "Log failure, skip runtime mode" [label="no"];
    "Start container with resource limits" -> "Verify container is running";
    "Verify container is running" -> "Container running?";
    "Container running?" -> "Execute agent commands via docker exec" [label="yes"];
    "Container running?" -> "Log failure, skip runtime mode" [label="no"];
    "Execute agent commands via docker exec" -> "Stop and remove container";
    "Stop and remove container" -> "Lifecycle complete";
}

1. Build the Image

The Dockerfile is at workspace/raw/runtime/Dockerfile. It is generated based on target type:

Node.js CLI/Library:

FROM node:lts-slim
WORKDIR /app
COPY target/ /app/
RUN npm install --production 2>/dev/null || true
RUN npm link 2>/dev/null || true
RUN mkdir -p /output
ENTRYPOINT ["sleep", "infinity"]

Python:

FROM python:3.12-slim
WORKDIR /app
COPY target/ /app/
RUN pip install --no-cache-dir -r requirements.txt 2>/dev/null || true
RUN pip install --no-cache-dir -e . 2>/dev/null || true
RUN mkdir -p /output
ENTRYPOINT ["sleep", "infinity"]

Compiled Binary (Go, Rust, C):

FROM ubuntu:22.04
RUN apt-get update && apt-get install -y --no-install-recommends \
    ca-certificates file strace && rm -rf /var/lib/apt/lists/*
COPY target/binary /usr/local/bin/target
RUN chmod +x /usr/local/bin/target
RUN mkdir -p /output
ENTRYPOINT ["sleep", "infinity"]

Web Application:

FROM node:lts-slim
WORKDIR /app
COPY target/ /app/
RUN npm install --production 2>/dev/null || true
RUN mkdir -p /output
EXPOSE 3000
CMD ["npm", "start"]

Build command:

$RUNTIME build -t greenfield-${WORKSPACE}-target \
  -f workspace/raw/runtime/Dockerfile .

If the build fails, log the error to workspace/raw/runtime/build-log.txt and mark runtime mode as unavailable. The pipeline continues with other modes.

2. Start the Container

CLI/Library targets:

$RUNTIME run -d \
  --name greenfield-${WORKSPACE}-target \
  --memory=2g --cpus=2 --pids-limit=256 \
  --network=none --read-only \
  --tmpfs /tmp:rw,noexec,nosuid,size=256m \
  -v "$(pwd)/workspace/raw/runtime:/output:rw" \
  greenfield-${WORKSPACE}-target

Web application targets:

$RUNTIME run -d \
  --name greenfield-${WORKSPACE}-target \
  --memory=2g --cpus=2 --pids-limit=256 \
  --network=none \
  -p 127.0.0.1:3000:3000 \
  -v "$(pwd)/workspace/raw/runtime:/output:rw" \
  greenfield-${WORKSPACE}-target

3. Verify Running

$RUNTIME inspect --format='{{.State.Running}}' greenfield-${WORKSPACE}-target
# Expected: true

4. Cleanup

$RUNTIME stop --time=10 greenfield-${WORKSPACE}-target 2>/dev/null || true
$RUNTIME rm greenfield-${WORKSPACE}-target 2>/dev/null || true

Command Execution

All target interaction goes through docker exec (or podman exec).

Basic Command

timeout 30 $RUNTIME exec greenfield-${WORKSPACE}-target \
  sh -c 'command args 2>&1' \
  > workspace/raw/runtime/cli/output.txt

With Timeout Handling

timeout 30 $RUNTIME exec greenfield-${WORKSPACE}-target \
  sh -c 'command args 2>&1' > output.txt 2>&1

EXIT_CODE=$?
if [ $EXIT_CODE -eq 124 ]; then
  echo "TIMEOUT: Command killed after 30 seconds." >> output.txt
fi

With Environment Variables

$RUNTIME exec -e "DEBUG=true" -e "CONFIG_PATH=/app/config.json" \
  greenfield-${WORKSPACE}-target sh -c 'target-command 2>&1'

With Piped Input

echo "user input here" | \
  timeout 30 $RUNTIME exec -i greenfield-${WORKSPACE}-target \
  sh -c 'target-command' > output.txt 2>&1

Pre-Execution Checklist

Before executing commands, verify:

  1. Container is running: $RUNTIME inspect --format='{{.State.Running}}' greenfield-${WORKSPACE}-target
  2. Output directory is mounted: $RUNTIME exec greenfield-${WORKSPACE}-target test -d /output

Exploration Patterns

CLI Exploration

# Help and version discovery
timeout 30 $RUNTIME exec $CONTAINER sh -c 'target --help 2>&1' > cli/help.txt
timeout 30 $RUNTIME exec $CONTAINER sh -c 'target --version 2>&1' > cli/version.txt

# Subcommand enumeration: parse help output, try each subcommand
# Error exploration: invalid flags, missing args, bad input
# Config discovery: env vars, config files, default paths

Web Endpoint Discovery

# Probe root and common API paths
curl -s -D- http://127.0.0.1:3000/ > web/root-response.txt

for endpoint in /api /health /swagger.json /openapi.json; do
  STATUS=$(curl -s -o /dev/null -w '%{http_code}' "http://127.0.0.1:3000${endpoint}")
  echo "${endpoint} -> ${STATUS}" >> web/endpoint-scan.txt
done

File Extraction

# Copy a file from the container
$RUNTIME cp greenfield-${WORKSPACE}-target:/path/to/file \
  workspace/raw/runtime/extracted/

# List files inside the container
$RUNTIME exec greenfield-${WORKSPACE}-target find /app -type f -name '*.log' 2>/dev/null

Error Handling

Build failure: Log to build-log.txt, skip runtime mode, continue with static analysis.

Command timeout (exit 124): Capture partial output, note timeout, move to next command.

Container crash/OOM:

if [ "$($RUNTIME inspect --format='{{.State.Running}}' $CONTAINER 2>/dev/null)" != "true" ]; then
  $RUNTIME logs $CONTAINER > crash-log.txt 2>&1
  $RUNTIME start $CONTAINER  # Attempt restart
fi

Container errors are behavioral observations — document them as data, not just failures.

Resource Limits

ResourceDefaultNotes
Memory2 GB--memory=2g
CPU2 cores--cpus=2
Command timeout30 secondstimeout 30 on docker exec
PID limit256--pids-limit=256
NetworkNone--network=none (default)
GPUNeverNot granted
PrivilegedNeverNot granted

Security Restrictions (Non-Negotiable)

  • Never use --privileged
  • Never mount host paths outside workspace/raw/runtime/
  • Never bind ports to 0.0.0.0 (always 127.0.0.1)
  • Never grant GPU access
  • Never disable seccomp/AppArmor
  • All captured output goes to workspace/raw/runtime/
  • Avoid including raw credential values in captured output (use judgment, not regex)

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.