Skip to content

About

An AI Agent Skill for eCapture (旁观者)

Resources

Stars

4 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

eCapture AI Agent Skill

中文 · eCapture · Releases

An agent skill for using eCapture to capture TLS plaintext, reconstruct HTTP, export keys/pcapng, and inspect shell commands or SQL on Linux/Android without a CA certificate. The skill is intended for agent harnesses that load SKILL.md and its relative resources.

Updated on 2026-10-31, against eCapture v2.9.1, local source commit 5b664c9. Skill package version: 2.0.0. The installed binary's help and behavior take precedence when using another version.

What changed

  • Live HTTP/1.1 and HTTP/2 reconstruction with --text-source pcapng for TLS, GoTLS, and GnuTLS, plus optional raw packet/key artifacts.
  • Separate operational --logaddr and captured --eventaddr; raw TCP/WS output formats and typed eCaptureQ Protobuf integration.
  • Linux LoongArch64, current OpenSSL/GnuTLS mappings, Android 16/17 updates, cgroup filtering, perf reordering, and actual HTTP reload semantics.
  • Corrected Bash exit-code advice, Android module availability, default listener state, output examples, buffer units, and unsupported claims.
  • Release-metadata-based Linux installation with required SHA256 validation, including Loong64 checksums; root-free dry run and prerequisite checks.

Contents

Resource Purpose
SKILL.md Discovery metadata and core agent workflow
Quick reference Modules, flags, version mappings, diagnostics
Platform constraints Kernels, privileges, namespaces, release artifacts
Output and streaming Artifact formats, destinations, rotation, eCaptureQ
Runtime configuration Routes, JSON fields, reload limits
Source baseline Upstream changes and maintenance evidence
Command templates Adaptable capture recipes
scripts/ Installer, preflight, library discovery, shared helpers
tests/ Offline helper regression tests with synthetic releases/hosts

Keep the whole directory together when installing this skill in a harness; references and helper files are required. Follow that harness's skill-directory or package-import mechanism. Loading only the main file as a prompt omits its referenced procedures and does not provide executable tools or target access.

Target requirements

Linux x86_64 >= 4.18, aarch64 >= 5.5, or loongarch64 >= 6.6; Android x86_64/aarch64 only. Root or sufficient capabilities are needed for capture. Android builds provide tls, gotls, and bash; other modules are Linux-only. macOS/Windows can author this skill and inspect outputs; run eCapture on a supported target. PID/UID/cgroup filters may be ignored below kernel 5.2.

# On Linux, from the skill directory. Requires Bash, Python 3, curl or wget.
bash scripts/install-ecapture.sh --dry-run
sudo bash scripts/install-ecapture.sh v2.9.1
bash scripts/preflight-check.sh
bash scripts/detect-libs.sh --pid 1234

# Replace interface and executable with actual target values.
sudo ecapture tls -m text --text-source pcapng -i eth0 --eventaddr /tmp/http.txt
sudo ecapture gotls -e /usr/local/bin/caddy -m keylog --eventaddr /tmp/go.keys

Dry run fetches metadata only. Missing assets or SHA256 entries stop installation; no unverified fallback is used. --install-dir supports a writable user directory. Preflight exits 0 for basic prerequisites, 1 for a blocking issue, or 2 when non-root capabilities still need validation. It does not attach probes.

Validation

python3 -m unittest discover -s tests -v
for script in scripts/*.sh scripts/lib/*.sh; do bash -n "$script" || exit; done

Tests validate the skill helpers without network access, installation into system directories, or BPF. Real capture, eCapture builds/tests, and E2E require a supported Linux/Android environment. Captured data may contain secrets; keep it in the authorized destination and redact it in reports.

License

Apache-2.0, matching eCapture's license.

About

An AI Agent Skill for eCapture (旁观者)

Resources

Stars

4 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages