Extend Harness

What it does

When your agent uses it

• The user asks the running agent to gain a new capability — phrasings like build a tool
• Give yourself X
• Let me <do thing through you
• Anything that extends the agent itself at runtime

Example prompts

• “build a tool”
• “add a hook”
• “give yourself X”
• “/extend-harness”

Requirements

• Python 3
• A credential in MY_API_KEY

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

1. 1Decide scope (~/.harness/ for personal, /.harness/ for shared).
2. 2Write the executable with the right protocol — use write_file, not heredoc-via-bash, so the file is clean and easy to edit later.
3. 3chmod +x — non-executable files are silently ignored. This is the most common omission.
4. 4Verify by invoking the new extension as a real tool call on your next turn. Harness rediscovers tools every loop iteration, so the new…
5. 5Iterate on failure. If the result is an error or unexpected output, read it, edit the file with edit_file/write_file, and call again. Keep…

What it can do on your machine

Read from SKILL.md and the folder at commit 5f5d968. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Shell commands in SKILL.md call:

  • jq
  • curl

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Links to these hosts (documentation or services it may open):

  • agents.md

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Extend Harness loads about 2.4k tokens when it runs. Until then it costs about 175 tokens; SKILL.md has 923 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~175

When it runs · the whole SKILL.md, loaded when a task matches

~2.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from wedow/harness at commit 5f5d968, republished under its MIT licence (© wedow). 923 words, ~2,431 tokens.

name

extend-harness

description

Use whenever the user asks the running agent to gain a new capability — phrasings like "build a tool", "add a hook", "give yourself X", "let me <do thing> through you", or anything that extends the agent itself at runtime. Defines the plugin protocols, where to install them (.harness/ scopes), and the required test-through-tool-call iteration pattern. CRITICAL: "tool" here means a plugin the agent invokes via tool-calls — install it to a .harness/ directory; do NOT write a standalone CLI script in the working directory and do NOT modify the harness codebase under plugins/. Load this skill before acting on such a request rather than reverse-engineering the protocol from source.

commands/         # CLI subcommands (hs <name>)
tools/            # tools the model can call
hooks.d/<stage>/  # pipeline hooks; sorted by NN- prefix
providers/        # LLM API adapters (binary or .conf variant)
prompts/*.md      # appended to the system prompt
skills/<name>/    # on-demand instruction packs (SKILL.md)
plugins/<name>/   # plugin packs with the same layout

#!/usr/bin/env bash
set -euo pipefail
case "${1:-}" in
  --schema) cat <<'JSON'
{"name":"web_get","description":"GET a URL and return body",
 "input_schema":{"type":"object",
                 "properties":{"url":{"type":"string"}},
                 "required":["url"]}}
JSON
    ;;
  --describe) echo "GET a URL" ;;
  --exec)
    url="$(jq -r '.url' < /dev/stdin)"
    curl -fsS "$url"
    ;;
esac

start → assemble → send → receive → done
                    ↑                 │
                    │   tool_exec → tool_done
                    │                 │
                    └─────────────────┘

#!/usr/bin/env bash
# .harness/hooks.d/tool_exec/05-confirm
set -euo pipefail
tc="$(cat)"
[[ "$(echo "$tc" | jq -r '.name')" == "bash" ]] || { echo "$tc"; exit 0; }
cmd="$(echo "$tc" | jq -r '.input.command')"
read -p "run: $cmd ? [y/N] " r </dev/tty
[[ "$r" =~ ^[Yy] ]] || exit 1
echo "$tc"

#!/usr/bin/env bash
# .harness/hooks.d/receive/20-cost
set -euo pipefail
r="$(cat)"
in="$(echo "$r" | jq -r '.usage.input_tokens // 0')"
out="$(echo "$r" | jq -r '.usage.output_tokens // 0')"
echo "$(date -Iseconds) in=$in out=$out" >> "${HARNESS_SESSION}/cost.log"
echo "$r"   # pass through

.harness/skills/my-workflow/
  SKILL.md                # frontmatter + body
  references/details.md   # listed in <skill-resources>
  scripts/setup.sh        # listed in <skill-resources>

protocol=openai
description=My Endpoint
model=foo-1
url=https://api.example.com/v1/chat/completions
auth_env=MY_API_KEY

© wedow, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in plugins/core/skills/extend-harness of wedow/harness.

Open the folder on GitHubat commit 5f5d968