Hforge recursive investigate
Skill ldilov/harness-forge/skills/hforge-recursive-investigate
AI coding agent runtime that makes LLMs understand your codebase with repository-aware context, memory, and continuous learning
npx -y skills add ldilov/harness-forge --skill hforge-recursive-investigateAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
recursive investigation orchestration using harness forge. use when the task is hard, ambiguous, cross-module, or likely to benefit from Typed RLM, bounded subcalls, durable iterations, and replayable recursive artifacts instead of chat-only reasoning.
SKILL.md
2.7 KB, as published. Nobody here has run it
HForge Recursive Investigate
Trigger Signals
- the root cause is unclear and needs staged evidence gathering
- the task crosses multiple modules, services, or ownership boundaries
- the prompt is getting expensive and would benefit from compact root frames and durable artifacts
- the work needs Typed RLM, bounded code cells, proposal artifacts, or replayable investigation state
- a normal one-pass repo scan is likely to lose context or miss intermediate findings
Do not use this for straightforward single-file edits where ordinary repo inspection is enough.
Inspect First
AGENTS.md.hforge/agent-manifest.json.hforge/generated/agent-command-catalog.json.hforge/runtime/index.json.hforge/runtime/recursive/language-capabilities.json.hforge/runtime/recursive/runtime-inventory.json.hforge/runtime/recursive/escalation-heuristics.json.hforge/runtime/recursive/sessions/commands/hforge-recursive-investigate.md
Workflow
- confirm the workspace is initialized with
hforge status --root . --json - if runtime artifacts are missing, initialize first with
hforge bootstrap --root . --yes - inspect recursive support with
hforge recursive capabilities --root . --json - inspect host-runtime posture with
hforge recursive runtimes --root . --json - provision Python or PowerShell explicitly with
hforge recursive provision-runtime <python|powershell> --root . --jsononly when a workspace-managed alias is helpful - create a durable session with
hforge recursive plan "<objective>" --task-id <taskId> --root . --json - prefer
hforge recursive executewith a typed bundle before using freeform structured-analysis snippets - use
hforge recursive runonly for one bounded fallback step when Typed RLM is not the right fit - inspect
iterations,subcalls,cells,promotions,meta-ops,score, andreplayto summarize the work from durable artifacts - stay honest about support posture, especially for Cursor and OpenCode where recursive support is translated rather than native
Output Contract
A good recursive-investigate result should include:
- the reason recursive mode was chosen
- the session id and the posture discovered from recursive capabilities and runtime inventory
- whether the agent used Typed RLM, bounded structured analysis, or both
- the most relevant iteration, scorecard, replay, or proposal artifacts
- the next recommended action based on the durable evidence