statsclaw-codex

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Commit authorship policy: shipper and every other agent must NOT append any Co-authored-by trailer, tooling footer, or bot identity to commit messages. The user is the sole author of every commit produced by the framework.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Knowledge-sharing lifecycle: Opt-In → Acquire → Read → Extract → Consent → Upload. Drives when and how the distiller is dispatched, how brain-contributions.md is surfaced to the user, and how approved entries are PR-ed to statsclaw/brain-seedbank.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Core orchestration protocol for the StatsClaw 9-agent framework. Defines every gate, state transition, precondition, must-not rule, workflow catalog, and signal-handling policy that the leader enforces.

Enable, disable, or inspect the StatsClaw shared brain — the cross-user knowledge system that lets agents read curated statistical knowledge and optionally contribute back. Use this skill when the user says "turn on brain", "enable brain mode", "turn off brain", "check brain status", "brain on", "brain off", or invokes $brain with on/off/status. Saves the setting to context.md and clones the brain repos if enabling. Full contribution lifecycle lives in skills/brain-sync.

Extract reusable knowledge from the current session, privacy-scrub it, and (with explicit user consent) submit it to the shared StatsClaw brain-seedbank as a pull request. Use this skill when the user says "contribute", "share what I learned", "submit lessons", "add to brain", "contribute knowledge", or invokes $contribute. Bypasses the automatic frequency heuristic but applies the same quality gate and mandatory user-consent step — nothing is uploaded without explicit approval.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] GitHub credential detection and verification. Establishes whether gh CLI, SSH key, $GITHUB_TOKEN, or a credential helper is available before any network-touching work runs.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Artifact handoff protocol for the two-pipeline architecture. Defines which files are produced by whom, what each downstream teammate may read, and how the workspace repo receives the run log.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Two-pipeline isolation protocol. Specifies worktree-level (filesystem) and pipeline-level (information) isolation — which agents get which specs, how dispatch.sh --worktree creates and merges back, and what counts as a violation.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Full GitHub issue-patrol protocol: scan → triage → per-issue planner→builder→tester→scriber→reviewer→shipper chain → PR + reply → patrol-report.md. The user-facing entry is the `patrol` skill; this file contains the mechanics.

Run a prompt or another StatsClaw skill on a recurring interval (default every 10 minutes). Use this skill when the user says "every 5 minutes do X", "check the deploy every 10 min", "keep running patrol", "poll until Y", "monitor on an interval", or invokes $loop. Delegates to scripts/loop.sh which manages the recurring child process — never implements polling with sleep inside the main session.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Inter-agent mailbox protocol. Specifies how teammates drop HOLD_REQUEST / BLOCK / STATUS messages into the shared mailbox and how leader drains it on every transition.

Scan and auto-fix open GitHub issues across a target repository using the StatsClaw 9-agent workflow. Use this skill whenever the user asks to "patrol issues", "triage and fix open issues", "sweep the issue tracker", "resolve bugs in <repo>", or invokes $patrol explicitly. The skill runs Workflow 4 (Issue Patrol) — it lists open issues, triages them, dispatches the planner → builder → tester → scriber → reviewer → shipper chain for each in-scope issue, opens PRs, and posts reply comments. Accepts arguments of the form `<owner/repo> [base-branch]` or a GitHub URL.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] PII-removal protocol applied to every candidate brain entry before it leaves the user's machine. Four categories: identifiers, paths/references, code references (genericized), data references.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Automatic language-profile selection for a target repository by inspecting file markers (DESCRIPTION → r-package, pyproject.toml → python-package, etc.).

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Rendering of the visual progress bar that leader prints after every status.md update. Not triggered by user intent; purely a formatting protocol.

Run a convergence-check review of the current run without shipping — reviewer reads every artifact across code/test/simulation pipelines and issues a PASS / PASS WITH NOTE / STOP verdict. Use this skill when the user says "review this", "check the work", "audit the run", "give me a verdict", "is it ready to ship", or invokes $review. Runs Workflow 8 (Review Only) — reviewer only, no shipper dispatch.

Commit, push, and open a pull request for the reviewed changes in the current run — plus sync the workspace log and, if applicable, open a brain-seedbank PR. Use this skill whenever the user says "ship it", "push the changes", "open a PR", "commit and push", "send it to main", "release this", or invokes $ship-it. Runs Workflow 7 (Ship Only) — dispatches reviewer first if no recent review exists, aborts on STOP verdicts, and only ships on PASS or PASS WITH NOTE.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Workflow 10 mechanics. Defines the smallness criteria, the reduced dispatch chain, and the hard gates that remain (credentials, reviewer) even when the full two-pipeline architecture is collapsed.

Run a Monte Carlo simulation study to evaluate finite-sample properties (bias, RMSE, coverage, size, power) of a statistical estimator. Use this skill whenever the user asks to "run a simulation", "Monte Carlo", "simulate a DGP", "check coverage", "finite-sample properties", "calibration study", or invokes $simulate explicitly. Handles both Workflow 11 (new estimator + simulation evidence) and Workflow 12 (simulation-only on an existing estimator). Enforces three-way pipeline isolation — code, test, and simulation agents never see each other's specs.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Full Monte Carlo simulation protocol: DGP specification, harness design, convergence diagnostics, tolerance integrity, and coverage/size/power reporting. The user-facing entry is the `simulate` skill; this file contains the mechanics.

[Internal protocol — leader-only. Not a user-invocable skill; do NOT trigger from user input.] Centralized workflow-log repository sync. Governs acquisition of the workspace repo, runtime state layout, and the post-run push of logs and CHANGELOG/HANDOFF entries.