Skip to content

backlog.md

Latest commit

 

History

History
103 lines (78 loc) · 4.5 KB

backlog.md

File metadata and controls

103 lines (78 loc) · 4.5 KB
Raw

Feature Backlog

Prioritized backlog for work beyond v0.2.0-alpha.2. The strategic phase view lives in roadmap.md; this file is the working list. Items move from here into implementation, gated by the standard quality gates (release-policy.md).

Positioning constraint for every item: BBT stays a local, read-only Linux observation substrate. No remediation, no network, no daemons.

P1 — Deepen existing commands

The current surface is broad enough; the next value is depth.

host snapshot: process-pressure summary

  • Total/running/zombie process counts, top RSS consumers.
  • Reuses bbt-proc collection; new process_pressure block in host-snapshot.v1 (additive, optional field).
  • Agent facts: zombie_process_count, top_rss_process at host level.

host snapshot: failed-service summary

  • Where systemd is available: failed-unit count and names via read-only systemctl --failed --output=json (or show per unit).
  • Structured warning (not failure) on non-systemd hosts, matching the service inspect pattern.
  • Agent fact: failed_service_count with warning severity when nonzero.

proc inspect: open files and sockets detail mode

  • Optional flag (e.g. --detail fds): open file descriptors, socket inodes correlated with /proc/net, working directory and executable presence checks (deleted-binary detection).
  • Degrades to structured warnings when /proc/<pid>/fd is unreadable.
  • Likely requires a v2 addition to proc-inspect schema or an additive optional block; prefer additive.

fs inspect: richer permission diagnostics

  • uid/gid name resolution (carry-over from roadmap Phase 2 leftover).
  • Owner/group/current-user evidence chain for permission-denied cases.
  • Broken-symlink diagnostics (target missing vs. loop vs. too many levels).
  • Parent-directory permission chain inspection.

P2 — Widen the command surface

The disk domain is now complete (usage, filesystems, pressure) — three verified verbs backed by three distinct data sources (recursive traversal, /proc/mounts + statvfs, and kernel PSI). bbt fs list PATH and bbt fs tree PATH have since shipped as well — the non-recursive ls counterpart with kind, size, owner, mode, and mtime per entry and a stable name-byte sort, and the bounded-depth tree with per-subtree recursive totals reusing the du-style --depth semantics from disk usage. Remaining items in rough priority order:

  1. bbt project inspect PATH — one-shot project context: VCS presence and status summary, language/build-system markers, largest artifacts. (Design doc first; scope creep risk is high.)

Each new command ships with: schema + validation case in scripts/validate_schemas.py (enforced — the script fails on uncovered schemas), command doc under docs/commands/, agent triage facts, and structured-warning degradation.

P3 — Ergonomics and distribution

  • Optional compatibility aliases: bls, bdu, bdf, bps (thin argv remaps, human mode only; JSON/agent consumers should use canonical commands).
  • Shell completions (clap_complete) shipped in release archives.
  • Man pages generated from clap definitions.
  • musl static builds as additional release targets (zero-dependency deploys onto minimal/container hosts).

P4 — Internal quality

  • getpwuid → getpwuid_r migration if any concurrency is introduced (documented constraint in security-model.md).
  • Property/fuzz tests for /proc parsers (net hex addresses, status fields, cmdline splitting) — these parse kernel-formatted, not attacker-controlled, data, but robustness is cheap insurance.
  • Golden-output snapshot tests for human mode to catch accidental wording drift.
  • Container-based CI job exercising a non-systemd environment end to end.

v1.0 gate

Schema freeze per release-policy.md: all bbt.*.v1 schemas become append-only. Before freezing:

  • P1 items landed (they may still add optional schema fields).
  • Cross-command consistency review: field naming, warning phrasing, severity conventions, evidence style.
  • One external consumer exercise: drive a real agent workflow against the JSON/agent contracts and fix what chafes.

Non-goals (permanent, from positioning)

Mutating/remediation commands, network features, daemon mode, fleet telemetry, SIEM/EDR/scanner ambitions, and AI chat features remain out of scope. Post-v1.0 discussions about a separate action layer must happen in a separate project, not BBT.