Development discipline for agents, activated with a repository link.
Give your agent the work and this repository:
Fix the checkout retry bug. Use https://github.com/LeiterConsulting/exec-ctrl
The default branch, main, carries the current v2 preview. See the
preview 5 release
for changes and verification. For reproducible adoption, give your agent the
version tag
and retrieve all instructions from that revision. The older v2 branch is historical;
use main for current updates. The preserved v1 guide remains available.
The agent reads the entry instructions, inspects your project, selects relevant controls, and carries them through implementation, verification and handoff. You do not need to pick modules, install a plugin, or create a document pack.
Read START_HERE.md and follow its activation sequence. Treat this repository as the method source and the user's workspace as the target, unless the request explicitly asks to improve exec-ctrl itself. Retrieve linked files from the same revision. If the host cannot fetch repository content, report that limit; do not claim the framework was loaded.
- Proportionate control: a small fix needs a short record; complex work gets explicit dependencies, risk, owners and release gates.
- Selective guidance: load relevant modules for policy, Git, security, functional correctness, troubleshooting, architecture, documentation, delivery, team systems and enterprise rules.
- Evidence before completion: connect requirements to implementation and verification; distinguish source inspection, tests, deployment and live results.
- Fits existing teams: reuse their instructions, issues, reviews and CI. Record policy sources and unresolved conflicts instead of inventing rules.
- Portable activation: plain Markdown works with any agent that can retrieve and follow it. Optional local adapters support future sessions.
- Checkable records: an optional Python helper routes declared task facts, generates pending records, adds enterprise gates and checks evidence-record structure without network access.
- Current-input evidence: explicit-file snapshots detect changed inputs before closure; required evidence kinds distinguish build, deployment and live stages.
- Session continuity: checkpoints preserve scope, decisions, failed hypotheses and the next action across interrupted work and changing context.
flowchart LR
A[Task + repo link] --> B[Inspect target and policy]
B --> C[Select controls by risk]
C --> D[Implement and verify]
D --> E[Review evidence and hand off]
E -->|new facts or failures| B
| Need | Entry |
|---|---|
| Use exec-ctrl now | Activation contract |
| Understand the rules | Operating model |
| Apply controls to daily work | Task playbooks |
| Resume or manage a long session | Session loop |
| Configure an IDE | Adapters and retrieval |
| Apply company policy | Enterprise rules |
| Fit existing issue/PR systems | Integration mappings |
| Use optional local checks | Tooling |
| See task walkthroughs | Examples |
| Upgrade an existing adoption | Migration |
| Assess readiness and next work | Review and roadmap |
| Track version changes | Changelog |
| Maintain the framework | Contributing |
2.0.0-preview.5 brings the controlled real-project feedback and disposable release/rollback acceptance into a versioned public preview. It checks artifact identity despite a healthy response, preserves required review holds, and rejects stale README/changelog versions during repository validation. It retains the compact activation entry, session controls, typed policy gates and offline, read-only helper. See preview 5 publication evidence and test coverage for current results and reproduction commands.
The Ping Monitor revisit exercised a real Go defect in an isolated worktree. The policy/release scenarios exercised actual disposable artifacts and deliberately held missing approval. These controlled cases inform the guidance; they do not establish independent IDE adoption.
The preview 3 evolution record preserves its 97-test hosted validation and controlled lab results. Those historical results are not proof of this revision.
Previous controlled lab results remain versioned evidence. Independent task-plus-link trials across IDEs and the stable v2 release gate remain open; see field trials.
A URL cannot force an IDE to retrieve instructions. exec-ctrl also cannot enforce permissions, branch protection or organizational policy by itself. Use the host's controls and CI for enforcement; the framework makes the agent's obligations and evidence explicit.
Existing v1 users can keep their workflows. The v1 guide, templates and historical records remain available as versioned reference.
exec-ctrl is available under the MIT license.