SECURE DEVELOPMENT PLATFORM
Guardrails and skills for AI coding agents.
[for Agents]
The plugin we build alongside our consulting practice. Supply-chain policy and attributed permissions that keep you safe from potentially destructive actions, plus hand-crafted skills for AI-enabled software development. New commands and hooks ship every month.
WHY SDP
Agents are shipping code. Nothing is watching the supply chain.
AI coding agents install dependencies, run shell commands, and rewrite files faster than any human review can keep up with. Most pipelines still give them all the same permissions as a human. One silent pnpm install, one quiet curl | sh, and a takeover package or unattributed permission is already inside the repo. SDP closes that gap. It surfaces what the agent is about to do, explains the risk, and gates the call when policy says no. The same plugin ships hand-crafted skills for AI-enabled software development, so guardrails and toolbox arrive in one install.
THREE PILLARS
Surface. Explain. Gate.
Supply-chain policy
Trust-aware installs that refuse to silently accept packages whose maintainer or signing posture has degraded since they were pinned. Takeovers and lost provenance surface as errors with context, not warnings buried in a log.
Hooks that block drift
Pre-edit, pre-bash, and pre-commit hooks enforce the rules that matter for your repo: banned imports, secrets in diffs, file ownership boundaries. The agent gets a real explanation; nothing slips through on a silent retry.
Attributed permissions
Every elevated permission carries an audit trail: who granted it, when, and scoped to what. No one-click "yes and don't ask again" that quietly carries into the next session or the next teammate.
CAUGHT IN THE WILD
Two recent catches, unedited.
⏺ Bash(pnpm install 2>&1) ⎿ Error: Exit code 1 ? Verifying lockfile against supply-chain policies (424 entries)... ✗ Lockfile failed supply-chain policy check (424 entries in 3.9s) [ERR_PNPM_TRUST_DOWNGRADE] 3 lockfile entries failed verification: eslint-import-resolver-typescript@3.10.1 High-risk trust downgrade semver@6.3.1 High-risk trust downgrade undici-types@6.21.0 High-risk trust downgrade The lockfile contains entries that the active policies reject. Inspect recent changes to pnpm-lock.yaml before trusting it. If expected, run "pnpm clean --lockfile" and then "pnpm install" to rebuild from a fresh resolution. Alternatively, relax the policy that flagged them. ⏺ Three more packages flagged by trustPolicy: no-downgrade, all legitimate transitive deps, but the pinned versions show maintainer/signing changes since they were pinned. Safe fix per the error: regenerate the lockfile. ⏺ Bash(npm view ... maintainers) ⎿ Waiting… ──────────────────────────────────────────────────────── Bash command npm view eslint-import-resolver-typescript@3.10.1 maintainers; \ npm view semver@6.3.1 maintainers; \ npm view undici-types@6.21.0 maintainers Check maintainers for the three flagged packages Do you want to proceed? ❯ 1. Yes 2. Yes, and don't ask again for: npm view * 3. No
SKILLS
Skills that ship in the same plugin.
SDP isn't only guardrails. The same plugin carries hand-crafted skills for AI-enabled software development. Three new ones just landed in v0.2.0, with more in every release.
Prompt context engineering
/wtlabs:improve-prompt rewrites a raw ask into a context-engineered prompt. It analyzes intent, structures the message into clear sections, and reaches for examples over rules. The agent gets a brief it can actually act on.
Read-only Kubernetes debugging
/wtlabs:debug-k8s gathers diagnostics through kubectl, flux, helm, istioctl, and cmctl, then writes a structured root-cause report. Nothing mutates the cluster. You get the analysis without giving the agent a knife.
Per-worktree workspace coloring
/wtlabs:theme picks a color palette and writes the worktree's VS Code workspace file. Multiple agent sessions across worktrees stay visually distinct, so you don't paste into the wrong window.
SHIPPING CADENCE
A lot more lands every month.
We rely on the Secure Development Platform to build our own consulting tools. Every new pattern we hit in client engagements becomes a policy, a hook, or a skill that ships back into the platform. Below is what is live today and what is moving through next.
TODAY
- Trust-policy pnpm installs (no-downgrade by default)
- Versioned permission template installer
- Permission inventory by scope
- Allowlist auto-tuning from transcripts
- Prompt context-engineering rewrite
- Read-only Kubernetes root-cause reports
- Per-worktree VS Code workspace theming
ON DECK
- Hook bundles for common drift patterns
- Attribution audit log + SIEM export
- Multi-repo policy distribution
- Org-wide trust posture dashboard