Hk local run spec

Skill deepklarity/harness-kit/.claude/skills/hk-local-run-spec

A kit for building with AI agents and also the engineering patterns around it.

Install

npx -y skills add deepklarity/harness-kit --skill hk-local-run-spec

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Plan and execute an odin spec. Handles working directory, auth, nested-session detection, and post-run diagnostics. Use whenever the user wants to run a spec, test a spec, or do a smoke test. Triggers on: 'run this spec', 'test this spec', 'odin plan', 'smoke test', or /hk-local-run-spec.

SKILL.md

3.6 KB, as published. Nobody here has run it

/hk-local-run-spec — Spec Runner

Run the full odin spec workflow: plan, review, execute. Handles the gotchas so you don't have to remember them.

Usage

/hk-local-run-spec ../sample_specs/poem_spec.md
/hk-local-run-spec ../sample_specs/poem_spec.md --quick
/hk-local-run-spec ../sample_specs/poem_spec.md --mock
/hk-local-run-spec --exec-only a1b2

Arguments

Parse $ARGUMENTS to extract:

spec_path (required unless --exec-only): path to the spec markdown file
--quick: use --quiet mode (non-interactive, spinner only)
--mock: pass --mock to exec (no backend writes)
--exec-only <task_id>: skip planning, just execute a specific task

The critical gotcha

Odin plan/exec invokes claude -p as a subprocess. It cannot run from inside another Claude Code session. Nested Claude Code calls fail silently or hang.

This skill detects the nested session and provides the user with copy-paste commands instead of trying to run them directly.

Execution

Step 1: Detect environment

Check if we're inside a Claude Code session:

# Claude Code sets this env var when running
echo "${CLAUDE_CODE_ENTRYPOINT:-not_set}"

If we're inside Claude Code (the var is set or we detect we're in an agent context), we cannot run odin plan/exec directly. Instead, provide copy-paste commands.

Step 2: Resolve paths

REPO_ROOT=$(git rev-parse --show-toplevel)
WORK_DIR="$REPO_ROOT/odin/temp_test_dir"

Check that temp_test_dir/ exists and has a .env:

ls "$WORK_DIR/.env" 2>/dev/null

If missing, tell the user:

temp_test_dir/ not found or missing .env.
Create it: cd odin && mkdir -p temp_test_dir && cp .env.example temp_test_dir/.env
Then edit temp_test_dir/.env with your ODIN_ADMIN_USER and ODIN_ADMIN_PASSWORD.

Step 3: Generate commands

Build the commands based on arguments:

Plan phase (skip if --exec-only):

cd "$WORK_DIR" && odin plan <spec_path> --quiet

If --quick was passed, use --quiet (which implies --auto). Otherwise default to --auto for non-interactive execution.

Review phase (skip if --exec-only):

cd "$WORK_DIR" && odin specs
cd "$WORK_DIR" && odin status

Exec phase:

cd "$WORK_DIR" && odin exec <task_id> [--mock]

Step 4: Output

Since we're almost always inside Claude Code (that's where this skill runs), output the commands for the user to copy-paste into a regular terminal:

## Spec Run Commands

Run these from a regular terminal (not inside Claude Code):

### 1. Plan
cd <WORK_DIR> && odin plan <spec_path> --auto

### 2. Review
cd <WORK_DIR> && odin specs
cd <WORK_DIR> && odin status

### 3. Execute (run for each task)
cd <WORK_DIR> && odin exec <task_id> [--mock]

### 4. Post-run diagnostic
cd <REPO_ROOT>/taskit/taskit-backend && python testing_tools/spec_trace.py <spec_id> --brief

If --exec-only was passed, skip the plan/review sections and only show the exec + diagnostic commands.

Step 5: Post-run diagnostic hint

Always remind the user to run the diagnostic after execution:

After execution completes, check results with:
  /hk-local-diagnose spec <spec_id> --brief

This connects the two skills into a natural workflow.