challenge
Run a challenger persona against the substance of any chain artefact — bare = persona roster
Install
npx skills add https://github.com/pche-broken-artist/forge-of-thought/tree/main/.claude/skills/challenge
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install pche-broken-artist-forge-of-thought@llmmart
git clone https://github.com/pche-broken-artist/forge-of-thought.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole pche-broken-artist/forge-of-thought collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Challenger personas live as .claude/agents/challenger-<persona>.md —
one isolated agent per persona, each defined by the blind spots it
exists to find, said in its description; the roster is the scan of
those files. Adding a persona means adding an agent file from
templates/challenger-definition.md; this command does not change (who creates
a persona, and when: CLAUDE.md, Isolated reviewers).
Shared behaviour: the contract skill named in each persona file's
front-matter (CLAUDE.md, Isolated reviewers). Like /check, the
persona agent writes nothing: it returns its report and this command
files it. This command chooses the persona, passes the target, files
the report and presents it.
Bare /challenge — the roster. List the available personas (scan
.claude/agents/challenger-*.md) and recommend which fits the
project's subject. A recommendation, never a gate.
/challenge <persona> [artefact|path] [slug] — run it. The target
is an artefact named as /forge names it, or any file of the project by
its path (CLAUDE.md, Isolated reviewers); how it narrows the run is the
contract's and the persona file's. Remake the project's ledger
first, python scripts/forge-ledger.py <slug>, so that the agent
reads a current file (SOL.0680); then invoke the challenger-<persona>
subagent on the project (infer it from context; if ambiguous, ask),
naming the target artefact. Pass only the project path and the target,
nothing else (CLAUDE.md, Isolated reviewers).
Best used before the next layer is first derived from the target — for the intent, before the first layer below it — while an accepted challenge is still cheap to absorb, and again after any major shift of direction. Running it on a near-final artefact is late but not useless.
When it returns:
- File the report. Write it into the project as
challenges/YYYY-MM-DD-challenge-<persona>.md: the agent's text word for word, its front-matter included, nothing added, dropped or reworded but the IDs: the next freeCHL.NNNNof the project's sequence, in tens, set at the head of each challenge; thenpython scripts/forge-report.py <path>; the printed name is the one the ledger cites. Thenpython scripts/forge-ledger.py <slug>: how the report becomes rows istemplates/ledger.md's (Challenges). - Present it to the principal: the overall read first, then each challenge compressed to two or three sentences. Do not editorialise and do not defend earlier drafting choices — if you disagree with a challenge, say so plainly and separately, marked as your own view.
- Answer the challenger's open questions where the answers exist in our conversation but not in the documents, and flag those to the principal: they usually mean something true is missing from the intent.
- End by offering a walkthrough of the challenges
(
.claude/skills/walkthrough/SKILL.md, the one owner of its shape and of the verdict words). Whatacceptwrites here: the challenge is mended through/forge <artefact>in the artefact it concerns, stateaccepted; where it is mended and when it goes above its layer: CLAUDE.md, Isolated reviewers. The other verdicts are the walkthrough's. If he declines the walkthrough, the challenges wait.
A rejected challenge is a normal, healthy outcome. So is a challenge that survives three rounds unresolved — park it and move on.
Files (forge-of-thought)
-
SKILL.md 3.5 KB
--- description: Run a challenger persona against the substance of any chain artefact — bare = persona roster argument-hint: "[persona] [artefact|path] [project-slug]" --- Challenger personas live as `.claude/agents/challenger-<persona>.md` — one isolated agent per persona, each defined by the blind spots it exists to find, said in its `description`; the roster is the scan of those files. Adding a persona means adding an agent file from `templates/challenger-definition.md`; this command does not change (who creates a persona, and when: CLAUDE.md, Isolated reviewers). Shared behaviour: the contract skill named in each persona file's front-matter (CLAUDE.md, Isolated reviewers). Like `/check`, the persona agent writes nothing: it returns its report and this command files it. This command chooses the persona, passes the target, files the report and presents it. **Bare `/challenge` — the roster.** List the available personas (scan `.claude/agents/challenger-*.md`) and recommend which fits the project's subject. A recommendation, never a gate. **`/challenge <persona> [artefact|path] [slug]` — run it.** The target is an artefact named as `/forge` names it, or any file of the project by its path (CLAUDE.md, Isolated reviewers); how it narrows the run is the contract's and the persona file's. Remake the project's ledger first, `python scripts/forge-ledger.py <slug>`, so that the agent reads a current file (SOL.0680); then invoke the `challenger-<persona>` subagent on the project (infer it from context; if ambiguous, ask), naming the target artefact. Pass only the project path and the target, nothing else (CLAUDE.md, Isolated reviewers). Best used **before the next layer is first derived from the target** — for the intent, before the first layer below it — while an accepted challenge is still cheap to absorb, and again after any major shift of direction. Running it on a near-final artefact is late but not useless. When it returns: 1. File the report. Write it into the project as `challenges/YYYY-MM-DD-challenge-<persona>.md`: the agent's text word for word, its front-matter included, nothing added, dropped or reworded but the IDs: the next free `CHL.NNNN` of the project's sequence, in tens, set at the head of each challenge; then `python scripts/forge-report.py <path>`; the printed name is the one the ledger cites. Then `python scripts/forge-ledger.py <slug>`: how the report becomes rows is `templates/ledger.md`'s (Challenges). 2. Present it to the principal: the overall read first, then each challenge compressed to two or three sentences. Do not editorialise and do not defend earlier drafting choices — if you disagree with a challenge, say so plainly and separately, marked as your own view. 3. Answer the challenger's open questions where the answers exist in our conversation but not in the documents, and flag those to the principal: they usually mean something true is missing from the intent. 4. End by offering a **walkthrough** of the challenges (`.claude/skills/walkthrough/SKILL.md`, the one owner of its shape and of the verdict words). What `accept` writes here: the challenge is mended through `/forge <artefact>` in the artefact it concerns, state `accepted`; where it is mended and when it goes above its layer: CLAUDE.md, Isolated reviewers. The other verdicts are the walkthrough's. If he declines the walkthrough, the challenges wait. A rejected challenge is a normal, healthy outcome. So is a challenge that survives three rounds unresolved — park it and move on.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.